Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Depth Vision-Based Assessment of Bone Marrow Mesenchymal Stem Cell Differentiation Capacity in Patients with Congenital Scoliosis

Depth Vision-Based Assessment of Bone Marrow Mesenchymal Stem Cell Differentiation Capacity in... Hindawi Journal of Healthcare Engineering Volume 2022, Article ID 4890008, 11 pages https://doi.org/10.1155/2022/4890008 Research Article Depth Vision-Based Assessment of Bone Marrow Mesenchymal Stem Cell Differentiation Capacity in Patients with Congenital Scoliosis Ning Liang, Qiwen Zhang, and Bin He e ird Affiliated Hospital of Zunyi Medical University (Zunyi First People’s Hospital), Zunyi, Guizhou 563000, China Correspondence should be addressed to Bin He; 171847206@masu.edu.cn Received 23 January 2022; Accepted 7 March 2022; Published 12 April 2022 Academic Editor: Deepak Kumar Jain Copyright © 2022 Ning Liang et al. .is is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Congenital scoliosis (CS) is a lateral curvature of one or more segments of the spine due to spinal dysplasia during fetal life. CS is clinically defined as a curvature of the spine >10 due to structural abnormalities of the vertebrae during the embryonic period. Its etiology is unknown, but recent studies suggest that it may be closely related to genetic factors, environmental factors, and developmental abnormalities. .e induction methods and modern applications of bone marrow MSCs provide a reference for in- depth human research on the induction of differentiation of bone marrow MSCs into osteoblasts. In this paper, by reviewing and organizing the literature on bone marrow MSCs, we summarized and analyzed the biological properties and preparation of bone marrow MSCs, the methods of inducing osteoblasts, the applications in tissue engineering bone, the problems faced, and the future research directions and proposed a method to assess the differentiation ability of bone marrow MSCs in patients with congenital scoliosis based on depth visual characteristics and the change of the method. .e method reveals and evaluates the multidirectional differentiation potential of bone marrow MSCs, which can be induced to differentiate into osteoblasts in vitro and can be used to construct bone tissue engineering scaffolds in vitro using tissue engineering techniques. Based on the properties of bone marrow MSCs, their application in congenital scoliosis patients for trauma repair, cell replacement therapy, hematopoietic support, and gene therapy is quite promising. It is necessary to carry out research on the mechanism of osteogenic differentiation of bone marrow MSCs to provide guidance and reference value for their induced differentiation into osteoblasts. comprehensive retrospective analysis of the clinical features 1. Introduction of CS combined with other systemic malformations was Congenital scoliosis (CS) is a three-dimensional deformity performed to explore the possible intrinsic link between the of the spine due to the abnormal development of the ver- concomitant malformations of each system. .e patho- tebral structures resulting in a coronal curvature of more genesis of CS is related to vertebral body formation, which than 10 . CS is often combined with rib and thoracic de- develops from embryonic somatic segments. Most scholars formities that affect the patient’s appearance and lead to agree that both genetic and environmental factors that can irreversible pulmonary impairment [1–3]. In addition, other affect the normal development of the somites are likely to systemic malformations, including spinal cord, heart, kid- contribute to CS. .e extensive and sequential involvement ney, and gastrointestinal system, can be observed in CS of the somites and subsequent development will result in patients. .e incidence of combined congenital defects of different types of vertebral abnormalities and corresponding other systems in CS patients is 40%–66%. .e current re- clinical manifestations. .e current research focuses on ports on concomitant malformations in CS are mostly environmental factors: maternal gestation, especially during limited to the incidence and classification of intraspinal and somite formation, may have an impact on gene signal ex- cardiac malformations, and the incidence of concomitant pression and epigenetics; genetic factors: based on the malformations reported in studies varies widely. A constancy of the location of specific genes encoded in 2 Journal of Healthcare Engineering SCD can be divided into four subtypes, all of which are vertebrates (e.g., vertebral body formation genes), identifi- cation of candidate genes by homozygosity of animal and autosomal monogenic recessive, with mutations in the DLL3 type. .ese genes are all related to the NOTCH human gene sequences, and population-based single nu- cleotide polymorphism (SNP). Nucleotide polymorphisms signaling pathway, which plays an important role in or- (SNPs) were screened in the population [4–7]. With regard ganogenesis, development, and apoptosis during the early to environmental factors, many physicochemical factors development of various tissues and organs. Osteogenesis were found to cause vertebral developmental malformations within the cartilage is significantly restricted [9–11]. .e in animals, such as high heat, organophosphorus pesticides, study of SCD, a genetic disease with spinal deformity, and drugs that can induce vertebral malformations in ro- provided important information on the genetic pattern of CS, suggesting that these genes play an important role in dents; NO can induce vertebral developmental malforma- tions in chickens. Risk factors for vertebral malformations in the normal development of the spine. .rough various sequencing tools such as whole-genome sequencing, humans have been reported: in a controlled analysis of 228 patients with CS between the ages of 0 and 50 years and 268 whole-genome exon sequencing, and whole-genome as- sociation analysis, combined with various model animal individuals with normal spines, risk factors were considered to include maternal insulin-dependent diabetes mellitus, experiments such as zebrafish model and point mutant valproic acid, alcohol, smoking, hyperthermia, twin preg- mice, researchers have screened some pathogenic genes nancies, artificially assisted reproductive techniques, and associated with CS pathogenesis. DLL3 was used as a in vitro fertilization. candidate gene, and sequencing of the DLL3 gene in 46 CS .e rise of assisted reproductive technology has led to patients identified a new highly conserved missense mu- an increase in the incidence of multiple pregnancies and an tation (S225 N). Validation experiments in a point mutant mouse model revealed that HES7, an effector gene of increasing emphasis on epigenetics (i.e., no changes in DNA sequence but changes in gene expression resulting in NOTCH, encodes a HES transcriptional repressor, and when a single dose of HES7 is insufficient, it can cause different phenotypes). .e genes involved in the typing of vertebral defects are shown in Figure 1. Because of this defects in embryonic somatic segment development by interfering with fibroblast growth factor (FGF) expression technique, methylation of nutrients or histone modifica- tions, for example, can be detected allowing epigenetics to and eventually cause CS malformations in mice. .e explain the occurrence of congenital malformations and WNT3A gene may also be associated with somite devel- some syndromes. .e detection of genetic predisposition opment and CS. .e WNT3A/β-catenin pathway regulates divides CS into syndromic and disseminated non- the expression of somite boundary determination genes syndromic types. Most CS tends to be disseminated, except MESP2 and Ripply2 by activating the activity of DLL1 and for the typical syndromes, which have a tendency to cluster TBX6, and ultimately determines the boundary formation during somite development. By whole-genome exome in families and are considered autosomal recessive by linkage analysis. A review of identical and heterozygous sequencing and comparative genomic hybridization microarray, the TBX6 gene was targeted for possible as- twins who developed CS was summarized through study reports of CS family lines, and it was concluded that sociation with CS. TBX6 gene, known as T-box6, is located disseminated CS tends to be a polygenic genetic disorder at 16p11. .e translation product is involved in mesoderm influenced by environmental, epigenetic influences. Dis- development as a transcription factor and transcriptionally semination leads to a lack of typical family lines making regulates the morphogenesis process. .e discovery of screening for single nucleotide polymorphisms more ap- TBX6 could explain the formation of CS in 11% of the cases propriate for CS studies. Genes associated with vertebral studied. development in mice have been elucidated [8], and with the Currently, autologous bone graft, allogeneic bone graft, completion of human exome and whole-genome se- and artificial bone substitute graft are the main treatments for bone defects, but all methods have limitations that lead to quencing, gene-disease association analysis has become a research hotspot, and more progress has been made in poor clinical outcomes. Autologous bone harvesting is limited and more invasive, which can easily lead to post- association analysis of SNPs at genetic loci for CS. .e SNPs loci associated with CS genetic susceptibility are: (1) operative infections and complications. Allogeneic bone syndromic type: SCD-associated loci include: DLL3, grafting, although not a limited source, carries a potential MESP2, LFNG, HES7, TBX6; STD-associated loci include: risk of immune response. .erefore, rapid and safe methods MESP2; (2) nonsyndromic type: PAX1 polymorphism is for bone defect repair have become a research direction in associated with CS genetic susceptibility; WNT3A poly- the field of bone injury in recent years [12]. BMSCs (bone morphism has insufficient evidence to be associated with marrow mesenchymal stem cells) originate from mesoder- mal cells and are multipotent stem cells with multispectral CS genetic susceptibility. Regarding the loci associated with CS genetic susceptibility in the Chinese Han population, differentiation, high proliferative capacity, and easy gene transfection, which can differentiate into osteoblasts, HES7, TBX6, and LMX1A were identified. Although CS has been suggested to be autosomal dominant or recessive, the chondrocytes, adipocytes, and neuronal cells under different conditions. BMSCs are easy to obtain, have low immuno- mode of inheritance is generally unclear, and the current mainstream research supports polygenic inheritance of CS. genicity, are easy to transfect, and can effectively suppress .e genetic pattern of spondylus costal dysostosis (SCD), a immune rejection after allogeneic transplantation by rare syndrome associated with CS, is better understood, as transfecting BMSCs with specific genes. .erefore, BMSCs Journal of Healthcare Engineering 3 Vertebral segmentation defect Single occurrence Multiple occurrence Localized involvement Involvement of <10 vertebrae Extensive involvement Localized involvement of Involvement of >10 vertebrae cervicothoracolumbar buttress Defined e.g.Alagille: VATER: VA Defined Defined Undefined Undefined Undefined CTERL: Goldenhar: CH e.g. SCD (1, 2, 3, 4); STD E.g., Alagille: VATER, etc. ARGE etc. Figure 1: Schematic diagram of the genes involved in the typing of vertebral defects. are important source cells for regenerative medicine and engineering research. Moreover, BMSCs can easily accept exogenous gene introduction and have promising applica- tissue engineering. BMSCs are multipotent stem cells that can differentiate into a variety of tissue cells, have weak tions in hematopoietic reconstruction, tissue repair, and immunogenicity, and can be easily transfected with exog- gene therapy, making them highly promising vector cells for enous genes, making them a hot vector cell for tissue en- gene therapy. Currently, the commonly used methods for gineering research and showing a bright future in the field of the isolation and purification of BMSCs include whole bone gene therapy and cell therapy. As ideal target cells for gene marrow applanation culture and density gradient centri- therapy, BMSCs have many biological properties such as fugation. While BMSCs are identified mainly with the help high metabolic viability, favorable recombinant protein of their surface antigenicity, BMSCs express a variety of secretion, multi-directional differentiation potential under surface proteins such as CD29 and CD44 after cell appo- sition attachment but do not express hematopoietic stem cell different induction conditions, and strong in vitro expansion ability. In order to promote the sustainable transformation surface markers such as CD14 and CD34. .ese surface of BMSCs into osteoblasts, it is necessary to maintain the antigens are immunologically specific, with low expression continuous action of osteogenic inducers outside [13]. .ey of the major histocompatibility complexes MHC I and MHC are currently used extensively for cell and tissue replacement II and no expression of major costimulatory molecules such therapy in a variety of diseases. However, with the con- as CD40, CD80, and CD86. .us, allogeneic transplantation tinuous research on BMSCs, it has been found that the of BMSCs does not cause rejection. .erefore, BMSCs have number and proliferation and differentiation potential of received increasing attention and become the main seed cells BMSCs decreases with age, limiting the wide application of for gene therapy, cell therapy, and tissue-engineering-related autologous BMSCs for transplantation. .erefore, the search research, and have good application prospects [15]. for new sources of MSCs is one of the hot spots of stem cell Visual depth feature extraction and disease diagnosis based on biomedical images have become an integral and research at home and abroad in recent years. BMSCs’ bi- ological characteristics are derived from bone marrow increasingly important part of healthcare. Magnetic reso- stroma and have the characteristics of mesenchymal and nance image (MRI), Positron emission tomography (PET), endothelial cells, capable of expressing a variety of antigens, Computer tomography (CT), cone beam CT, 3D ultra- including adhesion factors, growth factors, many receptors sound imaging, and other medical imaging technologies and integrins, but not their specific antigens. MScs play the are widely used in clinical examination, diagnosis, treat- role of supporting and nourishing hematopoietic cells. ment, and decision-making. .e cell morphology of BMSCs play a role in supporting and nourishing hemato- BMSCs under transmission electron microscopy is shown poietic cells, a primitive cell subpopulation of non- in Figure 2. In this paper, we propose a visual depth feature- based method to evaluate the differentiation capacity of hematopoietic origin with a highly self-replicating, multispectral differentiation potential. BMSCs have a strong bone marrow MSCs in patients with congenital scoliosis, which uses artificial intelligence deep learning methods to proliferative and multi-directional differentiation potential and play an important role in bone metabolism, with a analyze and process these large-scale medical image data strong osteogenic potential and maintain the ability to repair and provides scientific methods and advanced technologies bone necrosis [14]. BMSCs are ideal seed cells for bone tissue for screening, diagnosis, treatment planning, treatment 4 Journal of Healthcare Engineering image guidance, efficacy assessment, and differentiation capacity evaluation of bone marrow MSCs in patients with congenital scoliosis in clinical medicine. It is a major scientific problem and a key technology of cutting-edge medical imaging that needs to be solved in the field of medical image analysis. 2. Related Work 2.1. Congenital Spinal Deformity. Congenital vertebral de- formities are widely recognized by clinical practitioners through the scoliosis deformities they cause. Winter et al. based on radiographic typing is widely accepted and applied by scholars to classify CS into vertebral body formation disorders, vertebral segmentation defects, and mixed types Figure 2: Cell morphology of BMSCs under transmission electron [16, 17]. Combined with genetic typing with the develop- microscopy. ment of genetics, researchers recognized that CS is geneti- cally related, and in order to recognize uniformity, the observation and evaluation, the child can also be prepared International Federation of Spinal Deformity and Scoliosis for surgery accordingly [18]. Conservative treatment is proposed a guided typing system on congenital vertebral mainly divided into cast and brace correction, and halo ring defects, which proved to have high reliability and validity. continuous traction. .e second “China-US TSRH Or- .is typing first classifies congenital vertebral defects into thopedic Surgeon’s Association” has dedicated a series of single defects, multiple localized defects (<10 affected ver- lectures to halo ring and cast correction: halo ring traction tebrae), and multiple extensive defects (≥10 affected verte- can be equipped with mobile carts, play apparatus, and brae). Fractionation based on frontal and lateral radiographs even special traction beds, with electronic devices that can can evaluate the vertebral body and the vertebral arch, but it adjust and monitor the amount of traction force recorded. does not show well for the posterior spinal complex and It is believed that halo ring traction can be used at any age, severe and complex scoliosis. With the development of CT for any type of scoliosis, and even for patients who have and the popularity of 3D CT, 3D fractionation of CS was failed after surgery. In addition to contraindications (e.g., proposed. .e first step: the vertebral defects were deter- intramedullary occupancy and spinal stenosis), subjective mined on the basis of single and multiple occurrences to intolerance, these limitations include stent height that is determine whether the vertebral body was consistent with too high to pass indoors. Fletcher et al. used a series of casts the posterior relationship; the second step subdivided the on 29 patients with IS and CS over 50 , successfully delaying location and morphology of the formation of the disorder; the first surgery by an average of 39 months, avoiding the the third part looked at whether there was a segmentation need for propping of growth rods in 21 children, and in disorder, again subdivided according to the location, mor- follow-up of 11 children with involvement of multiple phology, and whether the anterior and posterior sides were consistent. Eventually, CS was divided into: (1) single segments confirmed the effectiveness of brace treatment. It is important to note that bracing or cast immobilization is consistent defects (hemivertebrae, cuneiform vertebrae, butterfly vertebrae, etc.); (2) multiple consistent defects not a lateral compression of the thorax; rotational com- pression is the key, as otherwise cardiopulmonary function (simultaneous appearance of hemivertebrae, cuneiform vertebrae, or butterfly vertebrae in adjacent or different and rib development are easily compromised. .e biggest benefit of conservative treatment is that it delays the timing locations); (3) complex inconsistent defects (mismatch of surgery and reduces the number of nonfusion surgeries complex, mixed complex): this type still needs further in- [19]. For patients who have poor cardiopulmonary func- depth differentiation; (4) subsegmental defects (can be tion or whose cardiopulmonary function is greatly affected subdivided according to location, direction, and rib rela- by plaster support fixation, halo ring traction is more tionship subdivision). However, there are no studies related recommended; conservative treatment cannot cure the to the etiology, embryology, and the expected progression of 3D subtypes. disease, and strict follow-up observation is needed during and after treatment. Progression of scoliosis tends to occur during peak growth (birth to late 4 years and adolescence), and CS is no exception. .e risk of progression is currently considered 2.2. Machine Learning Instructional Evaluation. Bone low for fully blocked vertebrae, cuneiform vertebrae, and marrow stromal cells are a heterogeneous population of unsegmented hemivertebrae; partial or fully segmented, many cells, among which exist mesenchymal stem cells, especially multiple hemivertebrae, are more likely to which are adult stem cells with self-renewal, replication, and progress; unilateral nonsegmented thoracic segments, es- multi-directional differentiation potential [20]. Bone mar- pecially combined with contralateral hemivertebrae, row MSCs can differentiate in two directions, osteogenesis progress rapidly, and there is a lack of systematic reports on and lipogenesis, and maintain a dynamic balance between the prognosis of complex mixed forms. After a period of Journal of Healthcare Engineering 5 capacity induces selective proliferation of BMSCs; stimulates the two. Once the physiological processes of osteogenesis and lipogenesis are disturbed, a number of metabolism- the expression of core binding factor α1 (RUNX2), alkaline phosphatase (ALP), osteopontin (OPN) and osteocalcin related and developmental disorders will result. In diseases such as osteoporosis and age-related bone loss, the reduction (OCN); and increases the mRNA expression level of ALP. of bone tissue is often accompanied by a significant accu- β-GP, as a source of phosphate in hydroxyapatite, provides mulation of adipose tissue. In some diseases characterized by phosphate ions, induces activation of ALP, and affects in- high bone mass, for example, increased abnormal bone tracellular signaling molecules. VitC is a cofactor of collagen formation and increased ossification are often accompanied prolyl hydroxylase, regulates extracellular matrix collagen by a decrease in intra-tissue fat content. .e effects of ab- homeostasis, and enhances DNA activity to promote cell differentiation. In recent years, several different classes of normal differentiation of BMSCs are particularly evident during the growth spurt. Due to significant changes in the substances have been reported to have the effect of pro- moting osteogenic differentiation of BMSCs in vitro, which amount of osteogenesis in adolescents, abnormal osteogenic differentiation can lead to a variety of clinical disorders, such can significantly improve the cellular purity of BMSCs [22]. Some herbal components have the effect of promoting os- as osteogenesis imperfecta, osseous malocclusion, and ab- normal bone age. In addition, trauma, severe infection, teogenic differentiation and proliferation of BMSCs, which tumor resection, or skeletal abnormalities can cause bone is beneficial for the repair of bone defects. BMSCs were defects, resulting in abnormal bone mass. .erefore, it is cultured in osteogenic induction medium containing Epi- important to explore the balance mechanism between os- medium as the experimental group and without Epimedium teogenic differentiation and lipogenic differentiation for the as the control group. .e results showed that the expression elucidation of the pathological mechanism and treatment of of osteogenic genes and the number of calcium nodules in the experimental group were significantly higher than those bone-related diseases as a clinical guide. .e source of BMSCs is limited, and the content of in the control group, confirming the synergistic effect of Epimedium and osteogenic induction medium in promoting BMSCs in bone marrow is very small, only 0.01%–0.1%, but BMSCs have strong proliferative ability in vitro, and the osteogenic differentiation. .e osteogenic induction me- dium of BMSCs was supplemented with different concen- number of passages has little effect on the proliferation ability of cells. .e morphology of primary and passaged tration gradients of osteopontin, and the results showed that bone marrow MSCs is shown in Figure 3. Currently, the osteopontin could promote osteogenic differentiation by international methods for in vitro isolation and culture of regulating the expression of RUNX2 and OCN proteins, thus BMSCs mainly include whole bone marrow apposition, contributing to the maintenance of the dynamic balance of density gradient centrifugation, flow cytometry sorting, and bone metabolism. immunomagnetic bead method. Among them, the most commonly used methods are whole bone marrow culture 2.3. Deep Vision Methods. Medical image analysis initially and density gradient centrifugation. .e flow cytometry focused on edge detection, texture features, morphological sorting method and immunomagnetic bead method use filtering, and the construction of shape models and template fluorescence and magnetic beads to label the surface antigens of BMSCs, which have the advantages of simpler operation, matching. .ese types of analysis methods are usually designed for specific tasks and are referred to as manual high sorting accuracy, and fast speed [21]. .e advantage of the density gradient centrifugation method is that it is easy bespoke design methods. Machine learning analyzes the task in a data-driven manner and can automatically learn rele- to operate, and the disadvantage is that it destroys the growth factors and the intact microenvironment of the vant model features and data characteristics from a large- scale dataset for a specific problem. Unlike models that are original BMSCs, which is not conducive to BMSCs colo- explicitly designed manually for a specific problem, machine nization. .e most commonly used methods are flow learning methods automatically learn medical image fea- cytometry and immunofluorescence staining, and changes tures implicitly and directly from data samples, and the in cell morphology can be observed using inverted mi- learning process is essentially an optimization problem- croscopy as an aid to identification. Flow cytometry: .ere solving process. .rough learning, the model selects the are mainly hematopoietic stem cells and BMSCs in the bone marrow, and BMSCs mainly express surface markers such as correct features from the training data, allowing the classifier to make the correct decisions when testing new data. CD44 and CD90, but not CD34 and CD45, which are surface markers of hematopoietic cells. Immunofluorescence .erefore, machine learning plays a crucial role in medical image analysis and has become the most promising area of staining method: By identifying the surface antigen ex- pression of BMSCs with immunofluorescence triple tech- research. Deep learning (DL) is a machine learning method that originated from the study of artificial neural networks nique, bone marrow mesenchymal cells were observed to and is motivated by the creation of neural networks that express CD29, CD90, and not CD45 under laser copolymer mimic the human brain to analyze and understand data. By microscopy. Induction of in vitro directed differentiation of observing how the visual center of the cat’s brain processes BMSCs into osteoblasts BMSCs in vitro osteogenic differ- retinotopic perceptual images, it was found that optic entiation relies heavily on osteogenic induction medium, neurons process information in a hierarchical manner, with which includes Dex, sodium β-glycerophosphate (β-GP), and vitamin C (VitC), which are essential cofactors for the different neurons focusing on different object features, and each layer of neurons abstracting some of the object features osteo-differentiation of BMSCs stem cells. Osteogenic 6 Journal of Healthcare Engineering a: Aer primary cultured for 3 d, cells were small b: Aer cultured for 7 d, the cells exhibited a and adherent, with a spindle-shaped appearance shuttle-shaped appearance c: Aer cultured for 2 wk, cells covered the whole d: Passage cells were stable and cell morphology bottom, exhibiting a spiral or swirling appearance did not change Figure 3: Morphology of primary and passaged bone marrow mesenchymal stem cells. for processing, with all information stimulated layer by layer, success in large vocabulary speech recognition systems, and the entire object perception stimulated at the highest reducing the speech recognition error rate by 30% relative to layer of the center. .e problem of gradient disappearance of the previous one. In its review, Harvard Medical School BP algorithm is solved by adopting unlabeled datasets for pointed out that the application of deep learning to solve network pretraining in deep feedforward networks. .e medical image analysis tasks is the development trend in this unsupervised greedy layer-by-layer training method is used field. In 2016, several experts have summarized, reviewed, to effectively reduce the dimensionality of the observed and discussed the current state of research and problems of objects, and then all network parameters are fine-tuned with deep learning in medical image analysis and the reviews supervised training. .is algorithm brings hope for solving published in relevant journals have summarized the research on deep learning in medical image classification, detection the deep structure-related optimization problem and has made a breakthrough in classification prediction such as and segmentation, and alignment and retrieval. .e classical image target recognition. Convolutional neural networks framework for CNN-based computer vision classification (CNNs), which use spatial relativity to reduce the number of tasks is shown in Table 1. parameters to improve training performance, were the first true multilayer structure learning algorithms. Long short- 3. Method term memory (LSTM) has also made breakthroughs in image handwriting recognition and speech recognition. 3.1.ModelArchitecture. Compared with traditional machine Deep learning has made important breakthroughs in several learning algorithms, visual deep learning algorithms avoid areas. In speech recognition, the introduction of Restricted the tedious image preprocessing and feature extraction steps Boltzmann machine (RBM) and Deep belief network (DBN) in traditional machine learning, and can be trained by di- into speech recognition model training has been a great rectly inputting images, which can preserve more features Journal of Healthcare Engineering 7 According to the role and function of each layer in CNN, and thus avoid errors and obtain higher recognition accu- racy. In this paper, we use Convolutional Neural Network they can be divided into the following five main types: input layer generally takes the vector form of a picture as input. (CNN) as a feedforward network, whose neurons can locally connect adjacent neurons and preserve the spatial structure Each layer input and output in FCN is a one-dimensional of the target through the special structure of the network, vector; and each layer input and output of CNN is arranged thus achieving a better performance in the field of image in three dimensions, similar to a rectangular body with processing. For CNN, its input layer can be directly input to three-dimensional dimensions of length, width, and depth. the original image, so it is widely used in many fields of In computer graphics, the depth of the rectangle reflects the computer vision, including image segmentation, image number of color channels in the image, which is 1 for classification, and image understanding. In this paper, we grayscale images and 3 for RGB images. .is part is called directly input the obtained BMSCs maximum profile “filter” (Filter). .e inner product of image and filter is the “convolution” operation, which is also the source of the grayscale images into the designed CNN for model learning and training, and then use the obtained models for classi- name of convolutional neural network. In the computation process, data with certain dimensions fication and recognition. .e overall architecture of the ∗ ∗ proposed method is shown in Figure 4. (width height depth) are used as input and convolved with the filter to obtain a two-dimensional array. .e process of filter forward propagation can be determined by setting the 3.2. Method Details. For a normal Fully Connected Neural size of the filter and the depth of the node matrix is obtained Network (FCN), it is basically the same as CNN in terms of by processing. Convolution kernel forward propagation is structure with input and output and training process. the process of getting the output of the next layer by the node Structurally, each node in CNN and FCN represents a output in the matrix of the previous layer through filter neuron, and in FCN, all nodes in two adjacent layers are action. Suppose the height, width, and depth of the input fully connected, while in CNN, nodes in two adjacent layers region are w,h,d, respectively, and for the i-th node in the are locally connected. .e input layer of CNN in image output unit node matrix, using bi to denote the corre- classification is the original picture of the image and the sponding bias in the i-th network node, the output value g(i) output layer represents the confidence level of different of the i-th node is classification categories, which is consistent with the output w h d of FCN. .e biggest problem of using FCN images is that i i ⎝ ⎠ ⎛ ⎞ g(i) � f 􏽘 􏽘 􏽘 a × w + b , (1) x,y,z x,y,z the fully connected layer has too many parameters, and x�1 y�1 z�1 they will cause the model to be computationally slow and prone to overfitting. .e CNN consists of two types of where x,y, and z are the values of the filter node (x,y,z) and f structures, one is the feature extraction layer, where is the activation function. .e source pixel is obtained as the neurons are connected to the local receptive fields of the unit target pixel of the rightmost matrix by the inner product upper layer neurons and can be used to extract local fea- operation after the filter action. tures, and each local feature corresponds to a network When a picture (grayscale image) with depth (Depth) of structure. .e other is the feature mapping layer, where only 1 is used as input, only the size of the filter needs to be each feature is considered as a plane with equal weights of set. .e input can be converted into a one-dimensional the mapping neurons. .e activation functions corre- vector expressing the tensor of the picture as the input of the sponding to the feature mapping structures are all dis- network. .e RGB color model is shown in Figure 5. h one- placement invariant. .e convolutional layer in CNN is dimensional vector input, and the set of target pixels cal- used for feature extraction, and this unique structure ef- culated by a filter is called the Feature Map, and the size of fectively reduces the feature resolution. Feature extraction the area mapped on the original image by the pixel points on process. At the same time, its structure has the feature of the feature map is called the Receptive Field. And, the weight sharing, so parallel learning is possible. .ese two common representation of the image is RGB color model features make CNNs structurally closer to real biological with 3 channels. .e following equation indicates the size neural systems, and they shine in the fields of natural and parameters of the filter, and r, g, and b in the lower language processing and image recognition. In addition, in corner indicate the weights of the red (red), green (green), practical applications, CNN can directly use the original and blue (blue) color channels, respectively. image as input, learn features by using small input data, and w w w w w w g1 g2 r1 r2 b1 b2 retain the spatial relationship between the lower pixels, so ⎣ ⎦ ⎡ ⎤ 􏼢 􏼣, , 􏼢 􏼣. (2) w w w w w w that objects in the image can still be detected by the net- g3 g4 r3 r4 b3 b4 work when scene migration or image transformation oc- .e filters then move up the input image in a certain curs. .erefore, convolutional neural networks have the order from the top left corner to the bottom right corner, following advantages: they can learn with fewer parameters and the distance of each move is called the stride. .e re- than fully connected neural networks; they can ignore the lationship between the size of the input matrix (input_size), effects of classifying and recognizing the location of objects the size of the filter (filter_size), and the size of the output in the image and image distortion; and they can auto- matrix (output_size) is satisfied as follows: matically learn and acquire features from the input data. 8 Journal of Healthcare Engineering Table 1: CNN-based classical framework for computer vision classification tasks. Network Features Remarks structure Multiple convolutional layers and subsampling LeNet American handwritten digit recognition layers Set a new world record in the ImageNet ILSVRC 2012 object AlexNet ReLU and dropout are proposed classification competition Proposed to use small convolution to verify Winner of ILSVRC 2014 for localization task and runner-up for VGGNet deeper networks and multi-scale fusion classification task 22-Layer network with multiple inception GoogleNet Winner of ILSVRC 2014 classification and detection task structures in series Proposed residual net, introduced jump Winner of the ILSVRC 2015 object detection and object recognition ResNet connection, 152 layers deep competition Inception Achieves comparable performance to ResNet, but with faster Inception structure combined with residual net ResNet convergence Densities prediction for pixel-level Avoids duplicate convolution computation due to overlap between FCN classification image blocks Mitigates gradient disappearance, enhances feature propagation, DenseNet Direct connection between any two layers supports feature reuse, and reduces the number of network parameters Simplify network structure and reduce network Achieve the same accuracy of AlexNet with only 1/50th of the number of SqueezeNet parameters AlexlNet parameters Proposed deformable deep convolutional DCNN Enhances the network’s ability to model geometric transformations neural network Combines the advantages of ResNet and .e DPN-based team won the 2017 ILSVRC object detection and object DPN DenseNet recognition competition Learn the importance of each feature channel SENet Winner of the 2017 ILSVRC image classification task competition and reinforce useful features Feature maps Output Input Convolutional layer Pooling layer Fully connected layer Figure 4: Model structure. 3.3. Data Enhancement. For the classification recognition output size − filter size (3) output size � + 1. model of convolutional neural networks, there is a con- stride clusion that 􏽲��������������������� h(log(2N/h) − log((η/4)) (4) P􏼠 test error≤ training error + 􏼡 � 1 − η. In the above equation, N is the number of training For the model complexity penalty, the smaller the h, the samples and h is the VC dimension of the classification smaller the penalty, and the larger the N, the smaller the model, where the root part is the model complexity penalty penalty. Deep learning models often have a large VC di- term. If the training model can make the training error rate mension and need a larger number of training samples to very low and the model complexity penalty term very low, reduce the penalty term; therefore, deep network structures the test error rate can be guaranteed to be at a very low level. like CNNs need large samples for model training to avoid Journal of Healthcare Engineering 9 height depth width Figure 5: RGB color model schematic. overfitting. Due to the high cost of obtaining confocal Import the InceptionV3 model and load the corre- sponding function modules. 2. Set the size of the imported microscopic images of BMSCs, the difficulty of related bi- ological experiments, and the limited number of images with images, the number of nodes in the fully connected layer, labels judged by combining protein expression determina- and the number of frozen layers; set the training set and tion and doctors’ clinical experience, a total of 128 samples validation set parameters and then use Image- are available, and data augmentation is needed for the DataGenerator() for data augmentation and image gen- BMSCs dataset. eration. 3. Fine-tuning (Fine-Tuning). After adding a new layer with the set number of classifications and base model, all previous layers are frozen and the correct 4. Experimentation and Evaluation bottleneck features are obtained to train the network layer by layer. 4. Classification and recognition. Load the 4.1. Dataset. In order to prevent the model from over- trained new model and perform classification and rec- fitting, data augmentation is needed for the obtained 128 ognition for the test set images. Among them, blue is the example samples. .e ImageDataGenerator() module in rising curve of the accuracy of the training set and green is keras can implement the basic data augmentation func- the rising curve of the accuracy of the test set. Finally, after tion. .is function can be used to generate augmented 10,000 iterations, the accuracy of the test set is 0.989, data cyclically during training until the iteration is which achieves an excellent classification and recognition completed. .e function has several operations that can be effect. used to augment the image data with different assign- ments. Seven data augmentation functions are selected, such as rotation, flip, translation, scale transformation, 4.3.ComparisonofResults. Table 2 shows the comparison of noise perturbation, color dithering, and contrast. .e 128 the classification recognition rates and time consumption of samples are augmented to 1200, and 900 are selected as the three classification recognition models for BMSCs, training samples and 300 as test samples. .e maximum showing that the convolutional neural network has a higher profile image of BMSCs is obtained as the input, and its accuracy rate for the classification recognition of images and resolution is 1000 ×1000. In order to reduce the model achieves the expected results of the experiments. computation and improve the computing efficiency, Gaussian Pyramid algorithm is used to downsample the image, so that its resolution is reduced to 64 × 64. .is set 4.4. Ablation Experiments. According to the experimental of images obtained is used as the input of BMSCs clas- results, the accuracy of classification recognition of sification recognition network. .e GoogleNet (Incep- BMSCs varies with the same batch size, learning rate, tionV3) model is borrowed and fine-tuned using Fine- activation function, and number of iterations due to the Tune to obtain the recognition accuracy of the model for different structure of the network, which proves that the BMSCs. deeper layers of the network have some advantages in the accuracy of recognition. .e MLP with 3 implied layers and 3,5,3 neurons per layer achieves the same highest 4.2. Experimental Steps. .e experimental hardware plat- classification rate as the MLP with 4 implied layers and form uses Inteli7-7700k processor, DDR4-16G memory, 3,3,5,5 neurons per layer, but due to its simpler model and GeForce1060 graphics card (CUDA acceleration with fewer training parameters, it is selected as the final module is available). .e implementation of InceptionV3 MLP structure for the classification of BMSCs. .e ac- classification and recognition algorithm is mainly based curacy of different network structures of MLP in iden- on the Keras library of Python platform, and the model is tifying normal and senescent cells in BMSCs is shown in trained using Fine-Tuning, and then the main process is as Table 3. follows: 1. Define the function and load the module. 10 Journal of Healthcare Engineering Table 2: Comparison of the recognition accuracy and time consumption of normal and aging BMSCs by different machine learning models. Model Svm MLP Inception V3 Average recognition rate 0.978 0.960 0.989 Time consuming 2.935s 1.762s 0.531s Table 3: Accuracy of normal versus senescent cell recognition in BMSCs with different network structure MLPs. ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ Network structure 3 3 5 3 3 5 3 3 3 3 5 3 3 3 5 3 3 5 5 Group 1 0.868 0.868 0.868 0.921 1.000 0.921 0.973 Group 2 0.842 0.895 0.868 0.973 0.921 1.000 0.947 Group 3 0.868 0.921 0.921 0.947 0.921 0.921 0.921 Group 4 0.895 0.895 0.921 0.973 1.000 1.000 1.000 Group 5 0.842 0.842 0.895 0.921 0.947 0.947 0.947 Group 6 0.895 0.921 0.895 1.000 0.973 0.947 0.973 Average recognition rate 0.868 0.890 0.895 0.956 0.960 0.956 0.960 5. Conclusion Conflicts of Interest .e incidence of intracanalicular deformity and osteo- .e authors declare that they have no conflicts of interest. chondral deformity differs significantly between types, and those with combined osteochondral deformity are more Acknowledgments likely to have intracanalicular deformity. Bone marrow mesenchymal stem cells have a wide range of clinical ap- .is work was sponsored in part by Fund Project approved by plications as stem cells that play an important role in Guizhou Provincial Department of Science and Technology supporting the proliferation of hematopoietic and hema- ([2018] 2756). topoietic progenitor cells, regulating the bone marrow mi- croenvironment, and have the potential for multi- References directional differentiation. Morphology-based taxonomic identification is of great significance for the evaluation of [1] J. Liu, N. Wu, N. Yang et al., “TBX6-associated congenital their physiological functions. In this paper, based on the scoliosis (TACS) as a clinically distinguishable subtype of optical sections of fluorescence confocal laser microscope, congenital scoliosis: further evidence supporting the com- morphological features were extracted, and various models pound inheritance and TBX6 gene dosage model,” Genetics in were designed on the Python platform using machine Medicine, vol. 21, no. 7, pp. 1548–1558, 2019. learning methods for algorithm development and validation. [2] C. E. Mackel, A. Jada, A. F. Samdani et al., “A comprehensive .e main contents include the preprocessing of optical review of the diagnosis and management of congenital sco- liosis,” Child’s Nervous System, vol. 34, no. 11, pp. 2155–2171, sections, the calculation of the maximum profile of the cell 3D structure, the calculation and selection of cell features, [3] M. Luo, W. Wang, N. Yang, and L. Xia, “Does three-di- and the design and implementation of machine learning mensional printing plus pedicle guider technology in severe models. In this paper, based on the selected cell features, we congenital scoliosis facilitate accurate and efficient pedicle propose an algorithm based on visual depth, use the data screw placement?” Clinical Orthopaedics and Related Re- augmentation method to realize the expansion of small search, vol. 477, no. 8, pp. 1904–1912, 2019. sample cell data, and use the migration learning method to [4] K. Takeda, I. Kou, S. Mizumoto et al., “Screening of known build a network structure suitable for the classification and disease genes in congenital scoliosis,” Molecular Genetics & recognition of this topic by InceptionV3 through the Fine- Genomic Medicine, vol. 6, no. 6, pp. 966–974, 2018. Tuning method to realize the classification and recognition, [5] W. Cho, N. Shepard, and V. Arlet, “.e etiology of congenital and the recognition accuracy is improved compared with the scoliosis: genetic vs. environmental—a report of three traditional machine learning method by up to 98.9%. monozygotic twin cases[J],” European Spine Journal, vol. 27, However, this algorithm learns a specific set of expressions no. 3, pp. 533–537, 2018. for a specific task, rather than learning a general model that [6] L. Wu, X. Zhang, Y. Wang, Y. Liu, and Y. Hai, “Risk factors for can be fully migrated. .erefore, in the future, we need to pulmonary complications after posterior spinal instrumen- study the generalization performance of the model to make tation and fusion in the treatment of congenital scoliosis: a case-control study[J],” BMC Musculoskeletal Disorders, the model have strong adaptability, so that it can be widely vol. 20, no. 1, pp. 1–8, 2019. used in a variety of diseases. [7] D. Hou, N. Kang, P. Yin, and Y. Hai, “Abnormalities asso- ciated with congenital scoliosis in high-altitude geographic Data Availability regions,” International Orthopaedics, vol. 42, no. 3, pp. 575–581, 2018. .e datasets used during the current study are available from [8] J. H Yang, D. G Chang, S. W Suh, W. Kim, and J. Park, the corresponding author on reasonable request. “Clinical and radiological outcomes of hemivertebra resection Journal of Healthcare Engineering 11 for congenital scoliosis in children under age 10 years: more than 5-year follow-up[J],” Medicine, vol. 99, no. 32, 2020. [9] Q. Zhao, B. Shi, X. Sun et al., “Do untreated intraspinal anomalies in congenital scoliosis impact the safety and effi- cacy of spinal correction surgery? A retrospective case-control study,” Journal of Neurosurgery: Spine, vol. 31, no. 1, pp. 40–45, 2019. [10] Q. J. Li, T. Yu, L. H. Liu, and J.-Wu Zhao, “Combined 3D rapid prototyping and computer navigation facilitate surgical treatment of congenital scoliosis: a case report and description of technique[J],” Medicine, vol. 97, no. 31, 2018. [11] K. Ahuja, S. Ifthekar, S. Mittal et al., “Is detethering necessary before deformity correction in congenital scoliosis associated with tethered cord syndrome: a meta-analysis of current evidence,” European Spine Journal, vol. 30, no. 3, pp. 599–611, [12] S. Mohamed-Ahmed, I. Fristad, S. A. Lie et al., “Adipose- derived and bone marrow mesenchymal stem cells: a donor- matched comparison[J],” Stem Cell Research & erapy, vol. 9, no. 1, pp. 1–15, 2018. [13] B. Wang, H. Wen, W. Smith, D. Hao, B. He, and L. Kong, “Regulation effects of melatonin on bone marrow mesen- chymal stem cell differentiation,” Journal of Cellular Physi- ology, vol. 234, no. 2, pp. 1008–1015, 2019. [14] H. Yu, J. Cheng, W. Shi et al., “Bone marrow mesenchymal stem cell-derived exosomes promote tendon regeneration by facilitating the proliferation and migration of endogenous tendon stem/progenitor cells,” Acta Biomaterialia, vol. 106, pp. 328–341, 2020. [15] L. Zhang, G. Jiao, S. Ren et al., “Exosomes from bone marrow mesenchymal stem cells enhance fracture healing through the promotion of osteogenesis and angiogenesis in a rat model of nonunion[J],” Stem Cell Research & erapy, vol. 11, no. 1, pp. 1–15, 2020. [16] Q. Sun, H. Nakata, M. Yamamoto, S. Kasugai, and S. Kuroda, “Comparison of gingiva-derived and bone marrow mesen- chymal stem cells for osteogenesis,” Journal of Cellular and Molecular Medicine, vol. 23, no. 11, pp. 7592–7601, 2019. [17] P. Li, D. Wang, L. Wang, and H. Lu, “Deep visual tracking: review and experimental comparison,” Pattern Recognition, vol. 76, pp. 323–338, 2018. [18] F. Xue, X. Wang, S. Li, Q. Wang, J. Wang, and H. Zha, “Beyond tracking: selecting memory and refining poses for deep visual odometry[C],” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 8575–8583, IEEE, Long Beach, CA, USA, 15 June 2019. [19] Y. Guo, Y. Li, L. Wang, and T. Rosing, “Depthwise convo- lution is all you need for learning multiple visual domains,” Proceedings of the AAAI Conference on Artificial Intelligence, vol. 33, no. 01, pp. 8368–8375, 2019. [20] B. Zhou, D. Bau, A. Oliva, and A. Torralba, “Interpreting deep visual representations via network dissection[J],” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 41, no. 9, pp. 2131–2145, 2018. [21] P. Gao, H. Li, S. Li et al., “Question-guided hybrid convolution for visual question answering[C],” in Proceedings of the Eu- ropean Conference on Computer Vision (ECCV), pp. 469–485, Springer, Munich, Germany, 8 September 2018. [22] Y. Wei, X. Wang, L. Nie, X. He, R. Hong, and T.-S. Chua, “MMGCN: multi-modal graph convolution network for personalized recommendation of micro-video[C],” in Pro- ceedings of the 27th ACM International Conference on Mul- timedia, pp. 1437–1445, ACM, Nice France, 21 October 2019. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal of Healthcare Engineering Hindawi Publishing Corporation

Depth Vision-Based Assessment of Bone Marrow Mesenchymal Stem Cell Differentiation Capacity in Patients with Congenital Scoliosis

Loading next page...
 
/lp/hindawi-publishing-corporation/depth-vision-based-assessment-of-bone-marrow-mesenchymal-stem-cell-ERgR2u5olF
Publisher
Hindawi Publishing Corporation
Copyright
Copyright © 2022 Ning Liang et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
ISSN
2040-2295
eISSN
2040-2309
DOI
10.1155/2022/4890008
Publisher site
See Article on Publisher Site

Abstract

Hindawi Journal of Healthcare Engineering Volume 2022, Article ID 4890008, 11 pages https://doi.org/10.1155/2022/4890008 Research Article Depth Vision-Based Assessment of Bone Marrow Mesenchymal Stem Cell Differentiation Capacity in Patients with Congenital Scoliosis Ning Liang, Qiwen Zhang, and Bin He e ird Affiliated Hospital of Zunyi Medical University (Zunyi First People’s Hospital), Zunyi, Guizhou 563000, China Correspondence should be addressed to Bin He; 171847206@masu.edu.cn Received 23 January 2022; Accepted 7 March 2022; Published 12 April 2022 Academic Editor: Deepak Kumar Jain Copyright © 2022 Ning Liang et al. .is is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Congenital scoliosis (CS) is a lateral curvature of one or more segments of the spine due to spinal dysplasia during fetal life. CS is clinically defined as a curvature of the spine >10 due to structural abnormalities of the vertebrae during the embryonic period. Its etiology is unknown, but recent studies suggest that it may be closely related to genetic factors, environmental factors, and developmental abnormalities. .e induction methods and modern applications of bone marrow MSCs provide a reference for in- depth human research on the induction of differentiation of bone marrow MSCs into osteoblasts. In this paper, by reviewing and organizing the literature on bone marrow MSCs, we summarized and analyzed the biological properties and preparation of bone marrow MSCs, the methods of inducing osteoblasts, the applications in tissue engineering bone, the problems faced, and the future research directions and proposed a method to assess the differentiation ability of bone marrow MSCs in patients with congenital scoliosis based on depth visual characteristics and the change of the method. .e method reveals and evaluates the multidirectional differentiation potential of bone marrow MSCs, which can be induced to differentiate into osteoblasts in vitro and can be used to construct bone tissue engineering scaffolds in vitro using tissue engineering techniques. Based on the properties of bone marrow MSCs, their application in congenital scoliosis patients for trauma repair, cell replacement therapy, hematopoietic support, and gene therapy is quite promising. It is necessary to carry out research on the mechanism of osteogenic differentiation of bone marrow MSCs to provide guidance and reference value for their induced differentiation into osteoblasts. comprehensive retrospective analysis of the clinical features 1. Introduction of CS combined with other systemic malformations was Congenital scoliosis (CS) is a three-dimensional deformity performed to explore the possible intrinsic link between the of the spine due to the abnormal development of the ver- concomitant malformations of each system. .e patho- tebral structures resulting in a coronal curvature of more genesis of CS is related to vertebral body formation, which than 10 . CS is often combined with rib and thoracic de- develops from embryonic somatic segments. Most scholars formities that affect the patient’s appearance and lead to agree that both genetic and environmental factors that can irreversible pulmonary impairment [1–3]. In addition, other affect the normal development of the somites are likely to systemic malformations, including spinal cord, heart, kid- contribute to CS. .e extensive and sequential involvement ney, and gastrointestinal system, can be observed in CS of the somites and subsequent development will result in patients. .e incidence of combined congenital defects of different types of vertebral abnormalities and corresponding other systems in CS patients is 40%–66%. .e current re- clinical manifestations. .e current research focuses on ports on concomitant malformations in CS are mostly environmental factors: maternal gestation, especially during limited to the incidence and classification of intraspinal and somite formation, may have an impact on gene signal ex- cardiac malformations, and the incidence of concomitant pression and epigenetics; genetic factors: based on the malformations reported in studies varies widely. A constancy of the location of specific genes encoded in 2 Journal of Healthcare Engineering SCD can be divided into four subtypes, all of which are vertebrates (e.g., vertebral body formation genes), identifi- cation of candidate genes by homozygosity of animal and autosomal monogenic recessive, with mutations in the DLL3 type. .ese genes are all related to the NOTCH human gene sequences, and population-based single nu- cleotide polymorphism (SNP). Nucleotide polymorphisms signaling pathway, which plays an important role in or- (SNPs) were screened in the population [4–7]. With regard ganogenesis, development, and apoptosis during the early to environmental factors, many physicochemical factors development of various tissues and organs. Osteogenesis were found to cause vertebral developmental malformations within the cartilage is significantly restricted [9–11]. .e in animals, such as high heat, organophosphorus pesticides, study of SCD, a genetic disease with spinal deformity, and drugs that can induce vertebral malformations in ro- provided important information on the genetic pattern of CS, suggesting that these genes play an important role in dents; NO can induce vertebral developmental malforma- tions in chickens. Risk factors for vertebral malformations in the normal development of the spine. .rough various sequencing tools such as whole-genome sequencing, humans have been reported: in a controlled analysis of 228 patients with CS between the ages of 0 and 50 years and 268 whole-genome exon sequencing, and whole-genome as- sociation analysis, combined with various model animal individuals with normal spines, risk factors were considered to include maternal insulin-dependent diabetes mellitus, experiments such as zebrafish model and point mutant valproic acid, alcohol, smoking, hyperthermia, twin preg- mice, researchers have screened some pathogenic genes nancies, artificially assisted reproductive techniques, and associated with CS pathogenesis. DLL3 was used as a in vitro fertilization. candidate gene, and sequencing of the DLL3 gene in 46 CS .e rise of assisted reproductive technology has led to patients identified a new highly conserved missense mu- an increase in the incidence of multiple pregnancies and an tation (S225 N). Validation experiments in a point mutant mouse model revealed that HES7, an effector gene of increasing emphasis on epigenetics (i.e., no changes in DNA sequence but changes in gene expression resulting in NOTCH, encodes a HES transcriptional repressor, and when a single dose of HES7 is insufficient, it can cause different phenotypes). .e genes involved in the typing of vertebral defects are shown in Figure 1. Because of this defects in embryonic somatic segment development by interfering with fibroblast growth factor (FGF) expression technique, methylation of nutrients or histone modifica- tions, for example, can be detected allowing epigenetics to and eventually cause CS malformations in mice. .e explain the occurrence of congenital malformations and WNT3A gene may also be associated with somite devel- some syndromes. .e detection of genetic predisposition opment and CS. .e WNT3A/β-catenin pathway regulates divides CS into syndromic and disseminated non- the expression of somite boundary determination genes syndromic types. Most CS tends to be disseminated, except MESP2 and Ripply2 by activating the activity of DLL1 and for the typical syndromes, which have a tendency to cluster TBX6, and ultimately determines the boundary formation during somite development. By whole-genome exome in families and are considered autosomal recessive by linkage analysis. A review of identical and heterozygous sequencing and comparative genomic hybridization microarray, the TBX6 gene was targeted for possible as- twins who developed CS was summarized through study reports of CS family lines, and it was concluded that sociation with CS. TBX6 gene, known as T-box6, is located disseminated CS tends to be a polygenic genetic disorder at 16p11. .e translation product is involved in mesoderm influenced by environmental, epigenetic influences. Dis- development as a transcription factor and transcriptionally semination leads to a lack of typical family lines making regulates the morphogenesis process. .e discovery of screening for single nucleotide polymorphisms more ap- TBX6 could explain the formation of CS in 11% of the cases propriate for CS studies. Genes associated with vertebral studied. development in mice have been elucidated [8], and with the Currently, autologous bone graft, allogeneic bone graft, completion of human exome and whole-genome se- and artificial bone substitute graft are the main treatments for bone defects, but all methods have limitations that lead to quencing, gene-disease association analysis has become a research hotspot, and more progress has been made in poor clinical outcomes. Autologous bone harvesting is limited and more invasive, which can easily lead to post- association analysis of SNPs at genetic loci for CS. .e SNPs loci associated with CS genetic susceptibility are: (1) operative infections and complications. Allogeneic bone syndromic type: SCD-associated loci include: DLL3, grafting, although not a limited source, carries a potential MESP2, LFNG, HES7, TBX6; STD-associated loci include: risk of immune response. .erefore, rapid and safe methods MESP2; (2) nonsyndromic type: PAX1 polymorphism is for bone defect repair have become a research direction in associated with CS genetic susceptibility; WNT3A poly- the field of bone injury in recent years [12]. BMSCs (bone morphism has insufficient evidence to be associated with marrow mesenchymal stem cells) originate from mesoder- mal cells and are multipotent stem cells with multispectral CS genetic susceptibility. Regarding the loci associated with CS genetic susceptibility in the Chinese Han population, differentiation, high proliferative capacity, and easy gene transfection, which can differentiate into osteoblasts, HES7, TBX6, and LMX1A were identified. Although CS has been suggested to be autosomal dominant or recessive, the chondrocytes, adipocytes, and neuronal cells under different conditions. BMSCs are easy to obtain, have low immuno- mode of inheritance is generally unclear, and the current mainstream research supports polygenic inheritance of CS. genicity, are easy to transfect, and can effectively suppress .e genetic pattern of spondylus costal dysostosis (SCD), a immune rejection after allogeneic transplantation by rare syndrome associated with CS, is better understood, as transfecting BMSCs with specific genes. .erefore, BMSCs Journal of Healthcare Engineering 3 Vertebral segmentation defect Single occurrence Multiple occurrence Localized involvement Involvement of <10 vertebrae Extensive involvement Localized involvement of Involvement of >10 vertebrae cervicothoracolumbar buttress Defined e.g.Alagille: VATER: VA Defined Defined Undefined Undefined Undefined CTERL: Goldenhar: CH e.g. SCD (1, 2, 3, 4); STD E.g., Alagille: VATER, etc. ARGE etc. Figure 1: Schematic diagram of the genes involved in the typing of vertebral defects. are important source cells for regenerative medicine and engineering research. Moreover, BMSCs can easily accept exogenous gene introduction and have promising applica- tissue engineering. BMSCs are multipotent stem cells that can differentiate into a variety of tissue cells, have weak tions in hematopoietic reconstruction, tissue repair, and immunogenicity, and can be easily transfected with exog- gene therapy, making them highly promising vector cells for enous genes, making them a hot vector cell for tissue en- gene therapy. Currently, the commonly used methods for gineering research and showing a bright future in the field of the isolation and purification of BMSCs include whole bone gene therapy and cell therapy. As ideal target cells for gene marrow applanation culture and density gradient centri- therapy, BMSCs have many biological properties such as fugation. While BMSCs are identified mainly with the help high metabolic viability, favorable recombinant protein of their surface antigenicity, BMSCs express a variety of secretion, multi-directional differentiation potential under surface proteins such as CD29 and CD44 after cell appo- sition attachment but do not express hematopoietic stem cell different induction conditions, and strong in vitro expansion ability. In order to promote the sustainable transformation surface markers such as CD14 and CD34. .ese surface of BMSCs into osteoblasts, it is necessary to maintain the antigens are immunologically specific, with low expression continuous action of osteogenic inducers outside [13]. .ey of the major histocompatibility complexes MHC I and MHC are currently used extensively for cell and tissue replacement II and no expression of major costimulatory molecules such therapy in a variety of diseases. However, with the con- as CD40, CD80, and CD86. .us, allogeneic transplantation tinuous research on BMSCs, it has been found that the of BMSCs does not cause rejection. .erefore, BMSCs have number and proliferation and differentiation potential of received increasing attention and become the main seed cells BMSCs decreases with age, limiting the wide application of for gene therapy, cell therapy, and tissue-engineering-related autologous BMSCs for transplantation. .erefore, the search research, and have good application prospects [15]. for new sources of MSCs is one of the hot spots of stem cell Visual depth feature extraction and disease diagnosis based on biomedical images have become an integral and research at home and abroad in recent years. BMSCs’ bi- ological characteristics are derived from bone marrow increasingly important part of healthcare. Magnetic reso- stroma and have the characteristics of mesenchymal and nance image (MRI), Positron emission tomography (PET), endothelial cells, capable of expressing a variety of antigens, Computer tomography (CT), cone beam CT, 3D ultra- including adhesion factors, growth factors, many receptors sound imaging, and other medical imaging technologies and integrins, but not their specific antigens. MScs play the are widely used in clinical examination, diagnosis, treat- role of supporting and nourishing hematopoietic cells. ment, and decision-making. .e cell morphology of BMSCs play a role in supporting and nourishing hemato- BMSCs under transmission electron microscopy is shown poietic cells, a primitive cell subpopulation of non- in Figure 2. In this paper, we propose a visual depth feature- based method to evaluate the differentiation capacity of hematopoietic origin with a highly self-replicating, multispectral differentiation potential. BMSCs have a strong bone marrow MSCs in patients with congenital scoliosis, which uses artificial intelligence deep learning methods to proliferative and multi-directional differentiation potential and play an important role in bone metabolism, with a analyze and process these large-scale medical image data strong osteogenic potential and maintain the ability to repair and provides scientific methods and advanced technologies bone necrosis [14]. BMSCs are ideal seed cells for bone tissue for screening, diagnosis, treatment planning, treatment 4 Journal of Healthcare Engineering image guidance, efficacy assessment, and differentiation capacity evaluation of bone marrow MSCs in patients with congenital scoliosis in clinical medicine. It is a major scientific problem and a key technology of cutting-edge medical imaging that needs to be solved in the field of medical image analysis. 2. Related Work 2.1. Congenital Spinal Deformity. Congenital vertebral de- formities are widely recognized by clinical practitioners through the scoliosis deformities they cause. Winter et al. based on radiographic typing is widely accepted and applied by scholars to classify CS into vertebral body formation disorders, vertebral segmentation defects, and mixed types Figure 2: Cell morphology of BMSCs under transmission electron [16, 17]. Combined with genetic typing with the develop- microscopy. ment of genetics, researchers recognized that CS is geneti- cally related, and in order to recognize uniformity, the observation and evaluation, the child can also be prepared International Federation of Spinal Deformity and Scoliosis for surgery accordingly [18]. Conservative treatment is proposed a guided typing system on congenital vertebral mainly divided into cast and brace correction, and halo ring defects, which proved to have high reliability and validity. continuous traction. .e second “China-US TSRH Or- .is typing first classifies congenital vertebral defects into thopedic Surgeon’s Association” has dedicated a series of single defects, multiple localized defects (<10 affected ver- lectures to halo ring and cast correction: halo ring traction tebrae), and multiple extensive defects (≥10 affected verte- can be equipped with mobile carts, play apparatus, and brae). Fractionation based on frontal and lateral radiographs even special traction beds, with electronic devices that can can evaluate the vertebral body and the vertebral arch, but it adjust and monitor the amount of traction force recorded. does not show well for the posterior spinal complex and It is believed that halo ring traction can be used at any age, severe and complex scoliosis. With the development of CT for any type of scoliosis, and even for patients who have and the popularity of 3D CT, 3D fractionation of CS was failed after surgery. In addition to contraindications (e.g., proposed. .e first step: the vertebral defects were deter- intramedullary occupancy and spinal stenosis), subjective mined on the basis of single and multiple occurrences to intolerance, these limitations include stent height that is determine whether the vertebral body was consistent with too high to pass indoors. Fletcher et al. used a series of casts the posterior relationship; the second step subdivided the on 29 patients with IS and CS over 50 , successfully delaying location and morphology of the formation of the disorder; the first surgery by an average of 39 months, avoiding the the third part looked at whether there was a segmentation need for propping of growth rods in 21 children, and in disorder, again subdivided according to the location, mor- follow-up of 11 children with involvement of multiple phology, and whether the anterior and posterior sides were consistent. Eventually, CS was divided into: (1) single segments confirmed the effectiveness of brace treatment. It is important to note that bracing or cast immobilization is consistent defects (hemivertebrae, cuneiform vertebrae, butterfly vertebrae, etc.); (2) multiple consistent defects not a lateral compression of the thorax; rotational com- pression is the key, as otherwise cardiopulmonary function (simultaneous appearance of hemivertebrae, cuneiform vertebrae, or butterfly vertebrae in adjacent or different and rib development are easily compromised. .e biggest benefit of conservative treatment is that it delays the timing locations); (3) complex inconsistent defects (mismatch of surgery and reduces the number of nonfusion surgeries complex, mixed complex): this type still needs further in- [19]. For patients who have poor cardiopulmonary func- depth differentiation; (4) subsegmental defects (can be tion or whose cardiopulmonary function is greatly affected subdivided according to location, direction, and rib rela- by plaster support fixation, halo ring traction is more tionship subdivision). However, there are no studies related recommended; conservative treatment cannot cure the to the etiology, embryology, and the expected progression of 3D subtypes. disease, and strict follow-up observation is needed during and after treatment. Progression of scoliosis tends to occur during peak growth (birth to late 4 years and adolescence), and CS is no exception. .e risk of progression is currently considered 2.2. Machine Learning Instructional Evaluation. Bone low for fully blocked vertebrae, cuneiform vertebrae, and marrow stromal cells are a heterogeneous population of unsegmented hemivertebrae; partial or fully segmented, many cells, among which exist mesenchymal stem cells, especially multiple hemivertebrae, are more likely to which are adult stem cells with self-renewal, replication, and progress; unilateral nonsegmented thoracic segments, es- multi-directional differentiation potential [20]. Bone mar- pecially combined with contralateral hemivertebrae, row MSCs can differentiate in two directions, osteogenesis progress rapidly, and there is a lack of systematic reports on and lipogenesis, and maintain a dynamic balance between the prognosis of complex mixed forms. After a period of Journal of Healthcare Engineering 5 capacity induces selective proliferation of BMSCs; stimulates the two. Once the physiological processes of osteogenesis and lipogenesis are disturbed, a number of metabolism- the expression of core binding factor α1 (RUNX2), alkaline phosphatase (ALP), osteopontin (OPN) and osteocalcin related and developmental disorders will result. In diseases such as osteoporosis and age-related bone loss, the reduction (OCN); and increases the mRNA expression level of ALP. of bone tissue is often accompanied by a significant accu- β-GP, as a source of phosphate in hydroxyapatite, provides mulation of adipose tissue. In some diseases characterized by phosphate ions, induces activation of ALP, and affects in- high bone mass, for example, increased abnormal bone tracellular signaling molecules. VitC is a cofactor of collagen formation and increased ossification are often accompanied prolyl hydroxylase, regulates extracellular matrix collagen by a decrease in intra-tissue fat content. .e effects of ab- homeostasis, and enhances DNA activity to promote cell differentiation. In recent years, several different classes of normal differentiation of BMSCs are particularly evident during the growth spurt. Due to significant changes in the substances have been reported to have the effect of pro- moting osteogenic differentiation of BMSCs in vitro, which amount of osteogenesis in adolescents, abnormal osteogenic differentiation can lead to a variety of clinical disorders, such can significantly improve the cellular purity of BMSCs [22]. Some herbal components have the effect of promoting os- as osteogenesis imperfecta, osseous malocclusion, and ab- normal bone age. In addition, trauma, severe infection, teogenic differentiation and proliferation of BMSCs, which tumor resection, or skeletal abnormalities can cause bone is beneficial for the repair of bone defects. BMSCs were defects, resulting in abnormal bone mass. .erefore, it is cultured in osteogenic induction medium containing Epi- important to explore the balance mechanism between os- medium as the experimental group and without Epimedium teogenic differentiation and lipogenic differentiation for the as the control group. .e results showed that the expression elucidation of the pathological mechanism and treatment of of osteogenic genes and the number of calcium nodules in the experimental group were significantly higher than those bone-related diseases as a clinical guide. .e source of BMSCs is limited, and the content of in the control group, confirming the synergistic effect of Epimedium and osteogenic induction medium in promoting BMSCs in bone marrow is very small, only 0.01%–0.1%, but BMSCs have strong proliferative ability in vitro, and the osteogenic differentiation. .e osteogenic induction me- dium of BMSCs was supplemented with different concen- number of passages has little effect on the proliferation ability of cells. .e morphology of primary and passaged tration gradients of osteopontin, and the results showed that bone marrow MSCs is shown in Figure 3. Currently, the osteopontin could promote osteogenic differentiation by international methods for in vitro isolation and culture of regulating the expression of RUNX2 and OCN proteins, thus BMSCs mainly include whole bone marrow apposition, contributing to the maintenance of the dynamic balance of density gradient centrifugation, flow cytometry sorting, and bone metabolism. immunomagnetic bead method. Among them, the most commonly used methods are whole bone marrow culture 2.3. Deep Vision Methods. Medical image analysis initially and density gradient centrifugation. .e flow cytometry focused on edge detection, texture features, morphological sorting method and immunomagnetic bead method use filtering, and the construction of shape models and template fluorescence and magnetic beads to label the surface antigens of BMSCs, which have the advantages of simpler operation, matching. .ese types of analysis methods are usually designed for specific tasks and are referred to as manual high sorting accuracy, and fast speed [21]. .e advantage of the density gradient centrifugation method is that it is easy bespoke design methods. Machine learning analyzes the task in a data-driven manner and can automatically learn rele- to operate, and the disadvantage is that it destroys the growth factors and the intact microenvironment of the vant model features and data characteristics from a large- scale dataset for a specific problem. Unlike models that are original BMSCs, which is not conducive to BMSCs colo- explicitly designed manually for a specific problem, machine nization. .e most commonly used methods are flow learning methods automatically learn medical image fea- cytometry and immunofluorescence staining, and changes tures implicitly and directly from data samples, and the in cell morphology can be observed using inverted mi- learning process is essentially an optimization problem- croscopy as an aid to identification. Flow cytometry: .ere solving process. .rough learning, the model selects the are mainly hematopoietic stem cells and BMSCs in the bone marrow, and BMSCs mainly express surface markers such as correct features from the training data, allowing the classifier to make the correct decisions when testing new data. CD44 and CD90, but not CD34 and CD45, which are surface markers of hematopoietic cells. Immunofluorescence .erefore, machine learning plays a crucial role in medical image analysis and has become the most promising area of staining method: By identifying the surface antigen ex- pression of BMSCs with immunofluorescence triple tech- research. Deep learning (DL) is a machine learning method that originated from the study of artificial neural networks nique, bone marrow mesenchymal cells were observed to and is motivated by the creation of neural networks that express CD29, CD90, and not CD45 under laser copolymer mimic the human brain to analyze and understand data. By microscopy. Induction of in vitro directed differentiation of observing how the visual center of the cat’s brain processes BMSCs into osteoblasts BMSCs in vitro osteogenic differ- retinotopic perceptual images, it was found that optic entiation relies heavily on osteogenic induction medium, neurons process information in a hierarchical manner, with which includes Dex, sodium β-glycerophosphate (β-GP), and vitamin C (VitC), which are essential cofactors for the different neurons focusing on different object features, and each layer of neurons abstracting some of the object features osteo-differentiation of BMSCs stem cells. Osteogenic 6 Journal of Healthcare Engineering a: Aer primary cultured for 3 d, cells were small b: Aer cultured for 7 d, the cells exhibited a and adherent, with a spindle-shaped appearance shuttle-shaped appearance c: Aer cultured for 2 wk, cells covered the whole d: Passage cells were stable and cell morphology bottom, exhibiting a spiral or swirling appearance did not change Figure 3: Morphology of primary and passaged bone marrow mesenchymal stem cells. for processing, with all information stimulated layer by layer, success in large vocabulary speech recognition systems, and the entire object perception stimulated at the highest reducing the speech recognition error rate by 30% relative to layer of the center. .e problem of gradient disappearance of the previous one. In its review, Harvard Medical School BP algorithm is solved by adopting unlabeled datasets for pointed out that the application of deep learning to solve network pretraining in deep feedforward networks. .e medical image analysis tasks is the development trend in this unsupervised greedy layer-by-layer training method is used field. In 2016, several experts have summarized, reviewed, to effectively reduce the dimensionality of the observed and discussed the current state of research and problems of objects, and then all network parameters are fine-tuned with deep learning in medical image analysis and the reviews supervised training. .is algorithm brings hope for solving published in relevant journals have summarized the research on deep learning in medical image classification, detection the deep structure-related optimization problem and has made a breakthrough in classification prediction such as and segmentation, and alignment and retrieval. .e classical image target recognition. Convolutional neural networks framework for CNN-based computer vision classification (CNNs), which use spatial relativity to reduce the number of tasks is shown in Table 1. parameters to improve training performance, were the first true multilayer structure learning algorithms. Long short- 3. Method term memory (LSTM) has also made breakthroughs in image handwriting recognition and speech recognition. 3.1.ModelArchitecture. Compared with traditional machine Deep learning has made important breakthroughs in several learning algorithms, visual deep learning algorithms avoid areas. In speech recognition, the introduction of Restricted the tedious image preprocessing and feature extraction steps Boltzmann machine (RBM) and Deep belief network (DBN) in traditional machine learning, and can be trained by di- into speech recognition model training has been a great rectly inputting images, which can preserve more features Journal of Healthcare Engineering 7 According to the role and function of each layer in CNN, and thus avoid errors and obtain higher recognition accu- racy. In this paper, we use Convolutional Neural Network they can be divided into the following five main types: input layer generally takes the vector form of a picture as input. (CNN) as a feedforward network, whose neurons can locally connect adjacent neurons and preserve the spatial structure Each layer input and output in FCN is a one-dimensional of the target through the special structure of the network, vector; and each layer input and output of CNN is arranged thus achieving a better performance in the field of image in three dimensions, similar to a rectangular body with processing. For CNN, its input layer can be directly input to three-dimensional dimensions of length, width, and depth. the original image, so it is widely used in many fields of In computer graphics, the depth of the rectangle reflects the computer vision, including image segmentation, image number of color channels in the image, which is 1 for classification, and image understanding. In this paper, we grayscale images and 3 for RGB images. .is part is called directly input the obtained BMSCs maximum profile “filter” (Filter). .e inner product of image and filter is the “convolution” operation, which is also the source of the grayscale images into the designed CNN for model learning and training, and then use the obtained models for classi- name of convolutional neural network. In the computation process, data with certain dimensions fication and recognition. .e overall architecture of the ∗ ∗ proposed method is shown in Figure 4. (width height depth) are used as input and convolved with the filter to obtain a two-dimensional array. .e process of filter forward propagation can be determined by setting the 3.2. Method Details. For a normal Fully Connected Neural size of the filter and the depth of the node matrix is obtained Network (FCN), it is basically the same as CNN in terms of by processing. Convolution kernel forward propagation is structure with input and output and training process. the process of getting the output of the next layer by the node Structurally, each node in CNN and FCN represents a output in the matrix of the previous layer through filter neuron, and in FCN, all nodes in two adjacent layers are action. Suppose the height, width, and depth of the input fully connected, while in CNN, nodes in two adjacent layers region are w,h,d, respectively, and for the i-th node in the are locally connected. .e input layer of CNN in image output unit node matrix, using bi to denote the corre- classification is the original picture of the image and the sponding bias in the i-th network node, the output value g(i) output layer represents the confidence level of different of the i-th node is classification categories, which is consistent with the output w h d of FCN. .e biggest problem of using FCN images is that i i ⎝ ⎠ ⎛ ⎞ g(i) � f 􏽘 􏽘 􏽘 a × w + b , (1) x,y,z x,y,z the fully connected layer has too many parameters, and x�1 y�1 z�1 they will cause the model to be computationally slow and prone to overfitting. .e CNN consists of two types of where x,y, and z are the values of the filter node (x,y,z) and f structures, one is the feature extraction layer, where is the activation function. .e source pixel is obtained as the neurons are connected to the local receptive fields of the unit target pixel of the rightmost matrix by the inner product upper layer neurons and can be used to extract local fea- operation after the filter action. tures, and each local feature corresponds to a network When a picture (grayscale image) with depth (Depth) of structure. .e other is the feature mapping layer, where only 1 is used as input, only the size of the filter needs to be each feature is considered as a plane with equal weights of set. .e input can be converted into a one-dimensional the mapping neurons. .e activation functions corre- vector expressing the tensor of the picture as the input of the sponding to the feature mapping structures are all dis- network. .e RGB color model is shown in Figure 5. h one- placement invariant. .e convolutional layer in CNN is dimensional vector input, and the set of target pixels cal- used for feature extraction, and this unique structure ef- culated by a filter is called the Feature Map, and the size of fectively reduces the feature resolution. Feature extraction the area mapped on the original image by the pixel points on process. At the same time, its structure has the feature of the feature map is called the Receptive Field. And, the weight sharing, so parallel learning is possible. .ese two common representation of the image is RGB color model features make CNNs structurally closer to real biological with 3 channels. .e following equation indicates the size neural systems, and they shine in the fields of natural and parameters of the filter, and r, g, and b in the lower language processing and image recognition. In addition, in corner indicate the weights of the red (red), green (green), practical applications, CNN can directly use the original and blue (blue) color channels, respectively. image as input, learn features by using small input data, and w w w w w w g1 g2 r1 r2 b1 b2 retain the spatial relationship between the lower pixels, so ⎣ ⎦ ⎡ ⎤ 􏼢 􏼣, , 􏼢 􏼣. (2) w w w w w w that objects in the image can still be detected by the net- g3 g4 r3 r4 b3 b4 work when scene migration or image transformation oc- .e filters then move up the input image in a certain curs. .erefore, convolutional neural networks have the order from the top left corner to the bottom right corner, following advantages: they can learn with fewer parameters and the distance of each move is called the stride. .e re- than fully connected neural networks; they can ignore the lationship between the size of the input matrix (input_size), effects of classifying and recognizing the location of objects the size of the filter (filter_size), and the size of the output in the image and image distortion; and they can auto- matrix (output_size) is satisfied as follows: matically learn and acquire features from the input data. 8 Journal of Healthcare Engineering Table 1: CNN-based classical framework for computer vision classification tasks. Network Features Remarks structure Multiple convolutional layers and subsampling LeNet American handwritten digit recognition layers Set a new world record in the ImageNet ILSVRC 2012 object AlexNet ReLU and dropout are proposed classification competition Proposed to use small convolution to verify Winner of ILSVRC 2014 for localization task and runner-up for VGGNet deeper networks and multi-scale fusion classification task 22-Layer network with multiple inception GoogleNet Winner of ILSVRC 2014 classification and detection task structures in series Proposed residual net, introduced jump Winner of the ILSVRC 2015 object detection and object recognition ResNet connection, 152 layers deep competition Inception Achieves comparable performance to ResNet, but with faster Inception structure combined with residual net ResNet convergence Densities prediction for pixel-level Avoids duplicate convolution computation due to overlap between FCN classification image blocks Mitigates gradient disappearance, enhances feature propagation, DenseNet Direct connection between any two layers supports feature reuse, and reduces the number of network parameters Simplify network structure and reduce network Achieve the same accuracy of AlexNet with only 1/50th of the number of SqueezeNet parameters AlexlNet parameters Proposed deformable deep convolutional DCNN Enhances the network’s ability to model geometric transformations neural network Combines the advantages of ResNet and .e DPN-based team won the 2017 ILSVRC object detection and object DPN DenseNet recognition competition Learn the importance of each feature channel SENet Winner of the 2017 ILSVRC image classification task competition and reinforce useful features Feature maps Output Input Convolutional layer Pooling layer Fully connected layer Figure 4: Model structure. 3.3. Data Enhancement. For the classification recognition output size − filter size (3) output size � + 1. model of convolutional neural networks, there is a con- stride clusion that 􏽲��������������������� h(log(2N/h) − log((η/4)) (4) P􏼠 test error≤ training error + 􏼡 � 1 − η. In the above equation, N is the number of training For the model complexity penalty, the smaller the h, the samples and h is the VC dimension of the classification smaller the penalty, and the larger the N, the smaller the model, where the root part is the model complexity penalty penalty. Deep learning models often have a large VC di- term. If the training model can make the training error rate mension and need a larger number of training samples to very low and the model complexity penalty term very low, reduce the penalty term; therefore, deep network structures the test error rate can be guaranteed to be at a very low level. like CNNs need large samples for model training to avoid Journal of Healthcare Engineering 9 height depth width Figure 5: RGB color model schematic. overfitting. Due to the high cost of obtaining confocal Import the InceptionV3 model and load the corre- sponding function modules. 2. Set the size of the imported microscopic images of BMSCs, the difficulty of related bi- ological experiments, and the limited number of images with images, the number of nodes in the fully connected layer, labels judged by combining protein expression determina- and the number of frozen layers; set the training set and tion and doctors’ clinical experience, a total of 128 samples validation set parameters and then use Image- are available, and data augmentation is needed for the DataGenerator() for data augmentation and image gen- BMSCs dataset. eration. 3. Fine-tuning (Fine-Tuning). After adding a new layer with the set number of classifications and base model, all previous layers are frozen and the correct 4. Experimentation and Evaluation bottleneck features are obtained to train the network layer by layer. 4. Classification and recognition. Load the 4.1. Dataset. In order to prevent the model from over- trained new model and perform classification and rec- fitting, data augmentation is needed for the obtained 128 ognition for the test set images. Among them, blue is the example samples. .e ImageDataGenerator() module in rising curve of the accuracy of the training set and green is keras can implement the basic data augmentation func- the rising curve of the accuracy of the test set. Finally, after tion. .is function can be used to generate augmented 10,000 iterations, the accuracy of the test set is 0.989, data cyclically during training until the iteration is which achieves an excellent classification and recognition completed. .e function has several operations that can be effect. used to augment the image data with different assign- ments. Seven data augmentation functions are selected, such as rotation, flip, translation, scale transformation, 4.3.ComparisonofResults. Table 2 shows the comparison of noise perturbation, color dithering, and contrast. .e 128 the classification recognition rates and time consumption of samples are augmented to 1200, and 900 are selected as the three classification recognition models for BMSCs, training samples and 300 as test samples. .e maximum showing that the convolutional neural network has a higher profile image of BMSCs is obtained as the input, and its accuracy rate for the classification recognition of images and resolution is 1000 ×1000. In order to reduce the model achieves the expected results of the experiments. computation and improve the computing efficiency, Gaussian Pyramid algorithm is used to downsample the image, so that its resolution is reduced to 64 × 64. .is set 4.4. Ablation Experiments. According to the experimental of images obtained is used as the input of BMSCs clas- results, the accuracy of classification recognition of sification recognition network. .e GoogleNet (Incep- BMSCs varies with the same batch size, learning rate, tionV3) model is borrowed and fine-tuned using Fine- activation function, and number of iterations due to the Tune to obtain the recognition accuracy of the model for different structure of the network, which proves that the BMSCs. deeper layers of the network have some advantages in the accuracy of recognition. .e MLP with 3 implied layers and 3,5,3 neurons per layer achieves the same highest 4.2. Experimental Steps. .e experimental hardware plat- classification rate as the MLP with 4 implied layers and form uses Inteli7-7700k processor, DDR4-16G memory, 3,3,5,5 neurons per layer, but due to its simpler model and GeForce1060 graphics card (CUDA acceleration with fewer training parameters, it is selected as the final module is available). .e implementation of InceptionV3 MLP structure for the classification of BMSCs. .e ac- classification and recognition algorithm is mainly based curacy of different network structures of MLP in iden- on the Keras library of Python platform, and the model is tifying normal and senescent cells in BMSCs is shown in trained using Fine-Tuning, and then the main process is as Table 3. follows: 1. Define the function and load the module. 10 Journal of Healthcare Engineering Table 2: Comparison of the recognition accuracy and time consumption of normal and aging BMSCs by different machine learning models. Model Svm MLP Inception V3 Average recognition rate 0.978 0.960 0.989 Time consuming 2.935s 1.762s 0.531s Table 3: Accuracy of normal versus senescent cell recognition in BMSCs with different network structure MLPs. ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ Network structure 3 3 5 3 3 5 3 3 3 3 5 3 3 3 5 3 3 5 5 Group 1 0.868 0.868 0.868 0.921 1.000 0.921 0.973 Group 2 0.842 0.895 0.868 0.973 0.921 1.000 0.947 Group 3 0.868 0.921 0.921 0.947 0.921 0.921 0.921 Group 4 0.895 0.895 0.921 0.973 1.000 1.000 1.000 Group 5 0.842 0.842 0.895 0.921 0.947 0.947 0.947 Group 6 0.895 0.921 0.895 1.000 0.973 0.947 0.973 Average recognition rate 0.868 0.890 0.895 0.956 0.960 0.956 0.960 5. Conclusion Conflicts of Interest .e incidence of intracanalicular deformity and osteo- .e authors declare that they have no conflicts of interest. chondral deformity differs significantly between types, and those with combined osteochondral deformity are more Acknowledgments likely to have intracanalicular deformity. Bone marrow mesenchymal stem cells have a wide range of clinical ap- .is work was sponsored in part by Fund Project approved by plications as stem cells that play an important role in Guizhou Provincial Department of Science and Technology supporting the proliferation of hematopoietic and hema- ([2018] 2756). topoietic progenitor cells, regulating the bone marrow mi- croenvironment, and have the potential for multi- References directional differentiation. Morphology-based taxonomic identification is of great significance for the evaluation of [1] J. Liu, N. Wu, N. Yang et al., “TBX6-associated congenital their physiological functions. In this paper, based on the scoliosis (TACS) as a clinically distinguishable subtype of optical sections of fluorescence confocal laser microscope, congenital scoliosis: further evidence supporting the com- morphological features were extracted, and various models pound inheritance and TBX6 gene dosage model,” Genetics in were designed on the Python platform using machine Medicine, vol. 21, no. 7, pp. 1548–1558, 2019. learning methods for algorithm development and validation. [2] C. E. Mackel, A. Jada, A. F. Samdani et al., “A comprehensive .e main contents include the preprocessing of optical review of the diagnosis and management of congenital sco- liosis,” Child’s Nervous System, vol. 34, no. 11, pp. 2155–2171, sections, the calculation of the maximum profile of the cell 3D structure, the calculation and selection of cell features, [3] M. Luo, W. Wang, N. Yang, and L. Xia, “Does three-di- and the design and implementation of machine learning mensional printing plus pedicle guider technology in severe models. In this paper, based on the selected cell features, we congenital scoliosis facilitate accurate and efficient pedicle propose an algorithm based on visual depth, use the data screw placement?” Clinical Orthopaedics and Related Re- augmentation method to realize the expansion of small search, vol. 477, no. 8, pp. 1904–1912, 2019. sample cell data, and use the migration learning method to [4] K. Takeda, I. Kou, S. Mizumoto et al., “Screening of known build a network structure suitable for the classification and disease genes in congenital scoliosis,” Molecular Genetics & recognition of this topic by InceptionV3 through the Fine- Genomic Medicine, vol. 6, no. 6, pp. 966–974, 2018. Tuning method to realize the classification and recognition, [5] W. Cho, N. Shepard, and V. Arlet, “.e etiology of congenital and the recognition accuracy is improved compared with the scoliosis: genetic vs. environmental—a report of three traditional machine learning method by up to 98.9%. monozygotic twin cases[J],” European Spine Journal, vol. 27, However, this algorithm learns a specific set of expressions no. 3, pp. 533–537, 2018. for a specific task, rather than learning a general model that [6] L. Wu, X. Zhang, Y. Wang, Y. Liu, and Y. Hai, “Risk factors for can be fully migrated. .erefore, in the future, we need to pulmonary complications after posterior spinal instrumen- study the generalization performance of the model to make tation and fusion in the treatment of congenital scoliosis: a case-control study[J],” BMC Musculoskeletal Disorders, the model have strong adaptability, so that it can be widely vol. 20, no. 1, pp. 1–8, 2019. used in a variety of diseases. [7] D. Hou, N. Kang, P. Yin, and Y. Hai, “Abnormalities asso- ciated with congenital scoliosis in high-altitude geographic Data Availability regions,” International Orthopaedics, vol. 42, no. 3, pp. 575–581, 2018. .e datasets used during the current study are available from [8] J. H Yang, D. G Chang, S. W Suh, W. Kim, and J. Park, the corresponding author on reasonable request. “Clinical and radiological outcomes of hemivertebra resection Journal of Healthcare Engineering 11 for congenital scoliosis in children under age 10 years: more than 5-year follow-up[J],” Medicine, vol. 99, no. 32, 2020. [9] Q. Zhao, B. Shi, X. Sun et al., “Do untreated intraspinal anomalies in congenital scoliosis impact the safety and effi- cacy of spinal correction surgery? A retrospective case-control study,” Journal of Neurosurgery: Spine, vol. 31, no. 1, pp. 40–45, 2019. [10] Q. J. Li, T. Yu, L. H. Liu, and J.-Wu Zhao, “Combined 3D rapid prototyping and computer navigation facilitate surgical treatment of congenital scoliosis: a case report and description of technique[J],” Medicine, vol. 97, no. 31, 2018. [11] K. Ahuja, S. Ifthekar, S. Mittal et al., “Is detethering necessary before deformity correction in congenital scoliosis associated with tethered cord syndrome: a meta-analysis of current evidence,” European Spine Journal, vol. 30, no. 3, pp. 599–611, [12] S. Mohamed-Ahmed, I. Fristad, S. A. Lie et al., “Adipose- derived and bone marrow mesenchymal stem cells: a donor- matched comparison[J],” Stem Cell Research & erapy, vol. 9, no. 1, pp. 1–15, 2018. [13] B. Wang, H. Wen, W. Smith, D. Hao, B. He, and L. Kong, “Regulation effects of melatonin on bone marrow mesen- chymal stem cell differentiation,” Journal of Cellular Physi- ology, vol. 234, no. 2, pp. 1008–1015, 2019. [14] H. Yu, J. Cheng, W. Shi et al., “Bone marrow mesenchymal stem cell-derived exosomes promote tendon regeneration by facilitating the proliferation and migration of endogenous tendon stem/progenitor cells,” Acta Biomaterialia, vol. 106, pp. 328–341, 2020. [15] L. Zhang, G. Jiao, S. Ren et al., “Exosomes from bone marrow mesenchymal stem cells enhance fracture healing through the promotion of osteogenesis and angiogenesis in a rat model of nonunion[J],” Stem Cell Research & erapy, vol. 11, no. 1, pp. 1–15, 2020. [16] Q. Sun, H. Nakata, M. Yamamoto, S. Kasugai, and S. Kuroda, “Comparison of gingiva-derived and bone marrow mesen- chymal stem cells for osteogenesis,” Journal of Cellular and Molecular Medicine, vol. 23, no. 11, pp. 7592–7601, 2019. [17] P. Li, D. Wang, L. Wang, and H. Lu, “Deep visual tracking: review and experimental comparison,” Pattern Recognition, vol. 76, pp. 323–338, 2018. [18] F. Xue, X. Wang, S. Li, Q. Wang, J. Wang, and H. Zha, “Beyond tracking: selecting memory and refining poses for deep visual odometry[C],” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 8575–8583, IEEE, Long Beach, CA, USA, 15 June 2019. [19] Y. Guo, Y. Li, L. Wang, and T. Rosing, “Depthwise convo- lution is all you need for learning multiple visual domains,” Proceedings of the AAAI Conference on Artificial Intelligence, vol. 33, no. 01, pp. 8368–8375, 2019. [20] B. Zhou, D. Bau, A. Oliva, and A. Torralba, “Interpreting deep visual representations via network dissection[J],” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 41, no. 9, pp. 2131–2145, 2018. [21] P. Gao, H. Li, S. Li et al., “Question-guided hybrid convolution for visual question answering[C],” in Proceedings of the Eu- ropean Conference on Computer Vision (ECCV), pp. 469–485, Springer, Munich, Germany, 8 September 2018. [22] Y. Wei, X. Wang, L. Nie, X. He, R. Hong, and T.-S. Chua, “MMGCN: multi-modal graph convolution network for personalized recommendation of micro-video[C],” in Pro- ceedings of the 27th ACM International Conference on Mul- timedia, pp. 1437–1445, ACM, Nice France, 21 October 2019.

Journal

Journal of Healthcare EngineeringHindawi Publishing Corporation

Published: Apr 12, 2022

References