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

Learn More →

Smart Healthcare System Based on Cloud-Internet of Things and Deep Learning

Smart Healthcare System Based on Cloud-Internet of Things and Deep Learning Hindawi Journal of Healthcare Engineering Volume 2021, Article ID 4109102, 10 pages https://doi.org/10.1155/2021/4109102 Research Article Smart Healthcare System Based on Cloud-Internet of Things and Deep Learning Benzhen Guo , Yanli Ma , Jingjing Yang , and Zhihui Wang College of Information Science and Engineering, Hebei North University, 11 Diamond South Road, Zhangjiakou 075000, China Correspondence should be addressed to Yanli Ma; hebeinu_ma@163.com Received 13 April 2021; Revised 31 May 2021; Accepted 20 June 2021; Published 29 June 2021 Academic Editor: Enas Abdulhay Copyright © 2021 Benzhen Guo 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. Introduction. Health monitoring and remote diagnosis can be realized through Smart Healthcare. In view of the existing problems such as simple measurement parameters of wearable devices, huge computing pressure of cloud servers, and lack of individ- ualization of diagnosis, a novel Cloud-Internet of )ings (C-IOT) framework for medical monitoring is put forward. Methods. Smart phones are adopted as gateway devices to achieve data standardization and preprocess to generate health gray-scale map uploaded to the cloud server. )e cloud server realizes the business logic processing and uses the deep learning model to carry out the gray-scale map calculation of health parameters. A deep learning model based on the convolution neural network (CNN) is constructed, in which six volunteers are selected to participate in the experiment, and their health data are marked by private doctors to generate initial data set. Results. Experimental results show the feasibility of the proposed framework. )e test data set is used to test the CNN model after training; the forecast accuracy is over 77.6%. Conclusion. )e CNN model performs well in the recognition of health status. Collectively, this Smart Healthcare System is expected to assist doctors by improving the diagnosis of health status in clinical practice. addition, the importance of feature learning is high- 1.Introduction lighted [7, 8]. )e feature representation of samples in the By the end of 2015, there were about 220 million people original space is transformed into a new feature space by aged 60 or above in China, accounting for 16.1% of the means of feature transformation layer by layer. Deep total population. Among them, 140 million were over 65 learning is widely used in image and voice processing and years old, accounting for 10.5% of the total population. In develops rapidly in the medical and health field [8–10]. this context, healthcare for the elderly has increasingly )e study in [11] proposed a human fall early warning aroused social concern [1]. )e rapid development of algorithm based on RNN and compared it with tradi- wearable devices, IOT, and cloud computing has brought tional machine learning algorithms such as SVM to verify about significant development opportunities and affected its excellent performance. )e study in [12] compared and studied the application of CNN, RNN, LSTM, and every aspect of people’s lives. In the field of healthcare, the adoption of IOT can bring us health monitoring all other deep learning models in the evaluation of sleep the time [2]. With the popularity of smart phones, smart quality. bracelets, and other devices, a variety of sensors can With the increase of the number of IOT devices and monitor health indicators timely and accurately [3–5]. sensors, medical IOT system also highlights more problems Deep learning is an implementation method of machine in the development process, mainly including the following: learning, which is different from the traditional shallow (1) )e measurement parameters of existing wearable models such as artificial neural network. Deep learning health monitoring devices are relatively simple, such usually has a deeper model structure with five or six layers as pedometers and intelligent sphygmomanometers. or even over ten layers or more hidden layers [6]. In )e common pedometer devices can connect with 2 Journal of Healthcare Engineering set. )e newly uploaded data are calculated by a smart phones via Bluetooth and other communi- cation means, and upload data to the cloud server, so depth model to automatically give health assessment results. )e model is a process of dynamic change. it is possible to view data such as exercise amount and sports assessment report through the phone )e private doctor will manually evaluate the user’s APPs. Various forms of wearable health monitoring health status and update the user data set on a regular devices fail to share data due to different manufac- basis as well as the parameter values of the depth turers and communication protocols. model. )e model can realize personalized health assessment. (2) Limited by memory capacity and computing power, wearable devices may upload the collected data to the cloud server through smart phones or home gate- 2. Materials and Methods ways. )e cloud server needs to store and process a 2.1. System Architecture. Figure 1 indicates the architecture large amount of collected data. )e server and of health monitoring system based on C-IOT and deep network have great transmission pressure, which learning proposed in this paper. It is improved from the reduces the real-time processing capacity. traditional three-layer architecture of the IOT, and described (3) At present, many telemedicine monitoring systems from the perspective of system implementation and data based on IOT carry out certain disease early warning flow, including data collection layer, data preprocessing and according to a set of diagnostic schemes, and it is net layer, data processing, and application layer. difficult to develop personalized diagnosis and Data collection layer is responsible for the acquisition of treatment schemes according to individual physio- physiological parameters of health, which is mainly com- logical characteristics and historical parameter posed of a number of Internet-connected or wearable de- changes. vices. Temperature, body fat/weight, blood pressure (4) Intelligent diagnosis methods based on the uploaded parameters, and exercise parameters are measured. In data of health monitoring devices mainly include general, data collection layer devices have small data storage system-based diagnosis methods of experts and in- capacity and low computing capacity [14–16]. In order to telligent diagnosis methods based on sample data. reduce data redundancy, network transmission pressure, Back Propagation (BP) neural network algorithm and power consumption, the data denoising, digital filtering, [13] and Support Vector Machines (SVM) algorithm and power management based on rules can be realized have been applied to the classification of diagnosis locally. High-frequency noise is eliminated by digital low- results. Although deep learning algorithm has been pass filter and band-pass filter. In addition, the special value applied in sleep quality and other fields [12], the in the collected data is removed based on rules. )e data application of intelligent diagnosis algorithm based collection layer device may be connected to smart phones or on deep learning is rarely seen. other mobile intelligent terminals through Bluetooth. In view of the above problems, a healthcare monitoring )e data preprocessing and net layer are mainly com- system based on Cloud-Internet of )ings (C-IOT) and deep posed of smart phones and other intelligent mobile termi- learning is proposed, and the major work includes the nals. Smart phone devices not only serve as network layer following aspects: devices to realize data communication function of medical IOT gateway, but also install application layer APP software (1) A health data acquisition system is designed based on to realize local data preprocessing, parameter display, and C-IOT and the acquisition of parameters such as device control. )e smart phone connects to the data col- human blood pressure, body temperature, body lection layer device through Bluetooth to receive mea- weight/fat, and exercise amount is realized. )e surement information reported by the blood pressure meter, shortcoming that the acquisition of simple physio- weight/fat meter, pedometer, thermometer, and other data logical parameters fails to evaluate and diagnose the acquisition devices, and display the real-time information on user’s health effectively is avoided. the mobile APPs. Preprocessing data are uploaded to the (2) Data preprocessing of each acquisition device is cloud server through Wifi, 4G, and other communication realized locally to eliminate noise interference. By means. Data preprocessing includes data normalization, communicating with the user’s smart phones data dimension transformation, data fusion, and other through Bluetooth, the smart phones may display the functions to generate images of health parameters. In terms collected data in real time, preprocess the data of of data normalization, body weight, body temperature, and each device, and upload them to the cloud server. other measurement parameters are normalized according to )e processing and integration of local data can the grade to the gray value ranging from 0 to 255. In terms of reduce the computing pressure of cloud server and data dimension transformation, original measurement pa- transmission pressure of network. rameters are extended in dimension, such as heart rate, (3) )e private doctor evaluates the user’s health con- systolic pressure and diastolic blood pressure measured by dition according to the data on the cloud server and sphygmomanometer extended to ambulatory pulse pressure establishes the initial data set. )e cloud server uses (APP), mean arterial pressure (MAP), and ambulatory rate- the deep learning algorithm (CNN) to train the user pressure product (ARPP), which are often used to diagnose health evaluation model according to the initial data cardiovascular diseases more effectively. )e application Journal of Healthcare Engineering 3 Elements Function Deep Data AI engine learning storage Data model processing and Private doctor training Intelligent Web application report Deep layer learning Model Server/database model update activition Users Router Data Data Data normalization transmission preprocessing Smartphone and net Data dimension Health image layer enhancement generation PAD Blue tooth Blood pressure Pedometer Data Data detection filtration Data collection layer Weight/fat Temperature Rule-based Power denoising management Figure 1: System architecture. software of mobile phones can also fuse the physiological trained deep learning model is used to solve the health pa- rameter image. With the increase of user data, private doctors parameter measurement data after normalization of gray value and dimension expansion into the image of health can annotate the data, enrich the personal health data set, and retrain and update the deep learning model. parameters, upload them to the cloud server in the form of two-dimensional image, and use the deep learning model to solve the health parameter image to give the health index 2.2. Measurement of Blood Pressure. For blood pressure report. )e users may receive and display the health report acquisition nodes, the oscillometric method is used to issued by the cloud server through smart phones. Making measure blood pressure [17], and the pressure value of the full use of the computing power of intelligent equipment for pressure sensor is filtered by low-pass filter and band-pass data preprocessing can effectively reduce the transmission filter to obtain the static pressure value and the dynamic pressure of database and network. With the rapid im- pressure value. In the measurement process, the dynamic provement of the computing power of intelligent terminal pressure value amplitude increases gradually and then de- creases. When the dynamic pressure value amplitude is equipment, more and more data processing functions will be completed directly in the intelligent terminals. multiplied by the normalized coefficients (Ks and Kd), the )e data processing and application layer mainly includes dynamic pressure value amplitude corresponding to the database and distributed server, web server, and deep learning systolic pressure and the diastolic pressure can be obtained, model engine. )e database stores and manages users’ per- respectively, and the human blood pressure value can be sonal information/health monitoring data, personal doctor obtained by reverse check, as shown in Figure 2. According information, equipment information, etc., the distributed to the conclusions of Mauro’s mathematical model, the server realizes various business processing logic, and the web normalized coefficient Ks was 0.46–0.64, and the normalized server provides users and personal doctors with friendly web coefficient Kd was 0.43–0.73 [18]. In this paper, the am- interface for background operation. )e deep learning model plitude coefficients commonly used in clinical medicine are engine is used to train the deep learning model, and the Ks � 0.48 and Kd � 0.58 [19]. 4 Journal of Healthcare Engineering fitting: a set of sample data (x , y )(i � 1, 2, 3 . . . N) can be i i At described by Gaussian functions as shown in the following equation: x − x 􏼁 i max (1) y � y ∗ exp􏼢− 􏼣. i max Kd = Ad/A max Ks = As/A max Take the logarithm of both sides of this equation: Ps Pc Pd Pt Figure 2: Blood pressure measurement method. In order to obtain the optimal pulse oscillation ampli- tude envelope, Gaussian fitting method is used for data 2 2 2 x − x 􏼁 x 2x x x i max max i max i (2) ln y � ln y − � 􏼠ln y − 􏼡 + − . i max max s s s s Assume preprocessing shall be conducted through the application software installed in the smart phone, which mainly includes ln y � z , i i dimensional transformation of data, data standardization, and generation of two-dimensional gray image for human max health. ln y − � b , max 0 In order to optimize the input vector, the measurement (3) data uploaded by monitoring equipment need to be 2x max � b , transformed into a certain dimension. )e measured data including heart rate (HR), systolic blood pressure (SP), and diastolic blood pressure (DP) are converted into three pa- − � b . rameters, namely, mean arterial pressure (MAP), ambula- tory pulse pressure (APP), and ambulatory rate-pressure Taking all sample data into consideration, equation (2) is product (ARPP). converted into a matrix as follows: APP � F1(SP, DP) � SP − DP, z 1 x x 1 1 ⎡ ⎢ ⎤ ⎥ ⎢ ⎥ ⎡ ⎢ ⎤ ⎥ ⎢ ⎥ b ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ 0 ⎢ ⎥ ⎢ ⎥ DP + (SP − DP) ⎢ ⎥ ⎢ 2 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ z ⎥ ⎢ ⎥⎡ ⎢ ⎤ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ 1 x x ⎥ (7) ⎢ 2 ⎥ ⎢ ⎢ ⎥ MAP � F2(SP, DP) � , ⎢ ⎥ ⎢ 2 2 ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ � ⎢ ⎥⎢ b ⎥. (4) ⎢ ⎥ ⎢ ⎥⎢ ⎥ 3 ⎢ ⎥ ⎢ ⎥⎢ 1 ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎣ ⎦ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢⋮ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢⋮ ⋮ ⋮ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎣ ⎦ ⎢ ⎥ ⎣ ⎦ 2 ARPP � F3(HR, SP) � HR∗ SP. 1 x x n n Step number, motion distance, fast motion time, fast It is simplified as motion distance, length of sleeping, and length of awakening Z � XB. (5) are obtained according to the step number and time mea- sured by the pedometer after dimension expansion. Data According to the principle of least squares, the gener- standardization converts the data of each index into the gray alized solution of least squares for matrix B is value ranged from 0 to 255 according to the scaling, so as to − 1 T T facilitate the synthesis of health status matrix for uploading (6) B � 􏼐X X􏼑 X Z. to the cloud server for processing. )e corresponding range relationship between the measured value of each index and )en, the estimated parameters (b , b , b ) are 0 1 2 the standardized value is shown in Table 1. substituted into equation (1) to obtain the fitted Gaussian In this paper, the deep learning method is adopted to function. Both the measuring speed and accuracy are im- evaluate the human health status. )e input data of the deep proved by using Gaussian fitting function. learning model are the measured values after the stan- dardization of each physiological parameter of human body in a day. In order to better serve as the input of CNN, the 2.3. Data Preprocessing. After the human physiological standardized data need to be processed in two dimensions to parameter monitoring equipment transmits the measure- generate the health status matrix. )e health status matrix is ment results to a smart phone via Bluetooth, data Journal of Healthcare Engineering 5 Table 1: Data standardization. Index Range of measurement value Range of standard value Weight 0–150 kg 0–255 Fat 0%–50% 0–255 SP 0–250 mmHg 0–255 DP 0–250 mmHg 0–255 HR 0–200 bpm 0–255 APP 0–200 mmHg 0–255 MAP 0–300 mmHg 0–255 ARPP 0–50000 0–255 Step num 0–30000 0–255 Distance 0–20 km 0–255 Fast motion time 0–24 h 0–255 Fast motion distance 0–20 km 0–255 Length of sleeping 0–24 h 0–255 Length of awakening 0–24 h 0–255 Body temperature 30–45 C 0–255 organized into 2D images. As required, the subjects shall 3 and 9 filters, respectively, and a scaling of 2 for the max- have their blood pressure and body temperature measured pooling layers. )e output layer outputs the health as- once every morning and every evening, have their weight/fat sessment results, including health, sub-health, and measured once every day, and wear a pedometer 24 hours a illness. day. On the smart phone side, 36-pixel 2D images of human health are generated every day, as shown in Table 2. Figure 3 3. Results and Discussion is an example 2D image of health status matrix according to 3.1. Measurement and Test of Blood Pressure. )e method of Table 2. using medical blood pressure meter and blood pressure measurement node at the same time for the same subject 2.4.CollectionofDataSet. )e training of deep learning model is used for comparative verification. A health monitor shall be supported by a certain amount of labeled data set. In this (PM-900S, Biocare Technology Co. Ltd., Shenzhen, study, the gray-scale chart of health monitoring parameters in China) is selected as the comparison device. )e verifi- one day is obtained by combining measurement and con- cation results are shown in Table 3, and the measurement struction. Six volunteers are selected, including 3 males and 3 results indicate that the relative error is within the range females (including 2 adolescents, 2 middle-aged, and 2 elderly). of 6%. Figure 5(a) is the photo of blood pressure mea- )e volunteers are required to test the blood pressure each surement node device. morning and evening, the body temperature once, and the Also, the method of wearing commercial pedometer weight/fat once. )ey are also required to wear a pedometer and step monitoring node is used for comparative ver- device for 24 hours and upload data to the cloud server at 6 ification. A sports bracelet (Honor 3, HUAWEI Tech- o’clock each morning through the phone APP. )en, private nology Co. LTD, Shenzhen, China) is selected as the doctors will grade their health status according to the uploaded comparison experiment device. )e test subjects wear data and user information such as age and gender and then sports bracelets and step monitoring nodes at the same divide them into three types: health, sub-health, and illness. )e time. )e measured data of the previous day is read after initial data set is generated by continuously tracking and an- the subjects get up at 6 o’clock in the morning. )e notating the user’s data for 100 days. In addition, the initial data verification results are shown in Table 4, and the mea- set of each user constructed and annotated by private doctors surement results indicate that the relative error is within according to the user’s historical data information is 3,000, and the range of 9%. Figure 5(b) is the photo of step moni- these three types account for 1/3. 80% of the extended initial toring node. data set is selected as the training sample to train the user’s personalized deep learning model, and 20% is used as the test sample to test the model’s performance. 3.2. Evaluation of Deep Learning Algorithm and Model. For each subject, the accuracy of average recognition of 2.5. Deep Learning Model. Deep learning model is con- the three health states is studied. Apart from that, recall structed and trained in the cloud server, and CNN is the and precision and F1_score are used for model evaluation most commonly used deep learning model, which has in respect of each category [21]. )e data of test set in data been widely used in the field of image processing. )e set are used as model input. TP denotes number of true classic LeNet-5 [20] CNN model is adopted and modified positive (labeled correctly). FP denotes number of false to simplify a pooling layer. )e CNN model constructed positive (other activity labeled as the sub-health and is shown in Figure 4, which contains two convolution illness). Furthermore, TN denotes number of true neg- layers, a pooling layer, and a full connection layer. )e atives (correct rejection), and FN denotes number of false CNNs have 2 × 2 kernels for the convolutional layers with negatives (missed detections). 6 Journal of Healthcare Engineering Table 2: Human health map. 1 2 3 4 5 6 1 SP1 DP1 HR1 APP1 MAP1 ARPP1 2 SP2 DP2 HR2 APP2 MAP2 ARPP2 3 Weight Fat 4 Step num Distance Fast motion time Fast motion distance Length of sleeping Length of awakening 5 Body temperature1 6 Body temperature2 0 12345 Figure 3: 2D images of human health. C1: 4 × 4@9 S1: 2 × 2@9 C1: 5 × 5@3 2 150 0 12345 Subsampling layer Max pooling: 2 × 2 Convolution layer Kernel size: 2 × 2 Convolution layer Full connection Somax Filter number: 3 Kernel size: 2 × 2 layer Stride length: 1 Filter number: 9 Stride length: 1 Figure 4: CNN model architecture. (S4) and the lowest is 68.5% (S3). )e standard deviation TP + TN Accuracy � , of recognition accuracy is all within 15 and the maximum TP + FP + FN + TN is 14.94 (S5). For 6 subjects, the average accuracy of CNN TP model is 77.61%. recall � , Figure 7 indicates the confusion matrices of 6 subjects TP + FN (8) acquired by CNN model. It can be revealed that the rec- TP ognition performance of the model for health categories is precision � , TP + FP obviously superior to the other two categories. In 200 samples, the highest recognition accuracy is 182 (S6). )e 2precision∗ recall F1 score � . capability to recognize sub-health categories is relatively precision + recall poor, and S5 can only recognize 110 out of 200 samples Figure 6 exhibits the accuracy of the average classi- correctly. fication of each subject. Error bars display the standard Figure 8 shows the precision, recall, and F1-score of 6 deviation of the recognition accuracy of the three health subjects in three categories acquired by CNN model. It categories for each subject. As can be seen from the figure, can be seen that CNN model boasts higher precision for the results of different subjects are of obvious differences. disease recognition than other two categories, except S2. )e highest recognition accuracy of CNN model is 84.2% Nevertheless, the value of recall is higher than that of Journal of Healthcare Engineering 7 Table 3: Blood pressure measurement verification results. SP (mmHg) (PM-900S) SP (mmHg) (test node) Error (%) DP (mmHg) (PM-900S) DP (mmHg) (test node) Error (%) 1 120 123 2.5 83 86 3.6 2 107 112 4.7 78 82 5.1 3 124 130 4.8 89 86 3.4 4 109 105 3.7 78 81 3.8 5 140 145 3.6 95 99 4.2 (a) (b) Figure 5: (a) Blood pressure measurement node device. (b) Step monitoring node device. Table 4: Step monitoring node verification results. Step num (honor3) Step num (test node) Error (%) Amount of sleep (h) (honor3) Amount of sleep (h) (test node) Error (%) 1 5600 5743 2.6 6.5 6.3 3.1 2 8212 7920 3.6 7.4 7.0 5.4 3 15401 16280 5.7 8.2 7.7 6.1 4 7231 7102 1.8 7.1 6.6 7.0 5 11005 10098 8.2 9.4 9.0 4.3 S1 S2 S3 S4 S5 S6 AVG CNN Figure 6: Classification accuracy of each subject. Each bar and the corresponding error bar show the average classification accuracy with standard deviation of three health patterns. Accuracy (%) 8 Journal of Healthcare Engineering Health Subhealth Illness Health Subhealth Illness Health Subhealth Illness 160 140 Health 171 18 11 Health 163 20 17 Health 150 34 16 140 120 Subhealth 30 152 18 Subhealth 22 140 38 Subhealth 46 120 34 Illness 922 169 Illness 10 12 178 Illness 25 34 141 20 20 (a) (b) (c) Health Subhealth Illness Health Subhealth Illness Health Subhealth Illness Health 180 12 8 Health 167 21 12 140 Health 182 11 7 100 100 Subhealth 28 153 19 Subhealth 49 110 41 Subhealth 32 145 23 60 60 40 40 40 Illness 12 16 172 Illness 21 25 154 Illness 33 20 147 20 20 (d) (e) (f) Figure 7: Confusion matrices acquired by the CNN model. (a–f) )e confusion matrix of S1–S6. 100 100 100 95 95 95 90 90 90 85 85 85 80 80 80 75 75 75 70 70 70 65 65 65 60 60 60 Health Subhealth Illness Health Subhealth Illness Health Subhealth Illness Precision (%) Precision (%) Precision (%) Recall (%) Recall (%) Recall (%) F1_score F1_score F1_score (a) (b) (c) 100 100 100 95 95 90 90 85 85 80 80 75 75 70 70 65 65 60 50 60 Health Subhealth Illness Health Subhealth Illness Health Subhealth Illness Precision (%) Precision (%) Precision (%) Recall (%) Recall (%) Recall (%) F1_score F1_score F1_score (d) (e) (f) Figure 8: Precision, recall, and F1-score obtained by the CNN model. (a–f) )e result curves of S1–S6. precision except S2. F1_score is a comprehensive index )e current research suggests that it is feasible to recognize and intelligently diagnose the preprocessed reflecting the performance of the model. By observing the F1_score curve of 6 subjects, the recognition performance data of the underlying devices of the medical Internet of of the model for sub-health is worse than the other two )ings with CNN model. In comparison with traditional categories, with a minimum of 61.8% (S5). machine learning methods such as SVM [22] and LDA Journal of Healthcare Engineering 9 [23], the depth learning method for intelligent diagnosis References does not require manual selection of data feature values. [1] S. Din and A. Paul, “Retracted: smart health monitoring and )e depth model can automatically extract features and management system: toward autonomous wearable sensing perform high-level abstraction. For different subjects, the for Internet of )ings using big data analytics,” Future optimal parameters of the model can be automatically Generation Computer Systems, vol. 91, pp. 611–619, 2019. acquired through training, which is more flexible and [2] W. Shi, J. Cao, Q. Zhang, Y. Li, and L. Xu, “Edge computing: robust compared with traditional machine learning vision and challenges,” IEEE Internet of 3ings Journal, vol. 3, methods. )e intelligent health management system ar- no. 5, pp. 637–646, 2016. chitecture based on deep learning proposed in the current [3] B. Kwolek and M. Kepski, “Human fall detection on em- research can make the data set continuously grow under bedded platform using depth maps and wireless accelerom- the condition of user data accumulation, and cooperate eter,” Computer Methods and Programs in Biomedicine, with doctors’ manual annotation, and can optimize vol. 117, no. 3, pp. 489–501, 2014. model training from time to time to achieve better rec- [4] Y. Kim and T. Moon, “Human detection and activity clas- sification based on micro-doppler signatures using deep ognition and diagnosis effects. Pretreatment of health convolutional neural networks,” IEEE Geoscience and Remote data at the bottom layer of Internet of )ings equipment Sensing Letters, vol. 13, no. 1, pp. 8–12, 2016. enhances the robustness of data and can tremendously [5] M. Z. Uddin, “A wearable sensor-based activity prediction lower the computing pressure of servers. system to facilitate edge computing in smart healthcare For each subject, the accuracy of average recognition of system,” Journal of Parallel and Distributed Computing, the three health states is studied. Apart from that, recall and vol. 123, pp. 46–53, 2019. precision and F1_score are used for model evaluation in [6] L. I. U. Dong, L. I. Yue, L. I. N. Jianping, L. I. Houqiang, and respect of each category. W. U. Feng, “Deep learning-based video coding: a review and a case study,” ACM Computing Surveys, vol. 53, no. 1, pp. 1–35, 4.Conclusions 2020. [7] M. Ghorbani, M. Bahaghighat, Q. Xin, and F. Ozen, “Con- In view of the existing problems in the existing IOT medical vLSTMConv network: a deep learning approach for sentiment system, a new IOT architecture for medical monitoring is analysis in cloud computing,” Journal of Cloud Computing, proposed in this paper, in which smart phones are used as vol. 9, no. 1, pp. 1–13, 2020. [8] L. Zheng, O. Wang, S. Hao et al., “Development of an early- gateway devices to realize data preprocessing of measurement warning system for high-risk patients for suicide attempt node devices, thus greatly reducing the computing pressure of using deep learning and electronic health records,” Transla- cloud servers and transmission pressure of network. Based on tional Psychiatry, vol. 10, no. 1, pp. 1–11, 2020. the data sets generated from such data annotated by private [9] Z. Liu, Y. Fu, Y. Zhang, F. Peng, and Z. Zhang, “Effect of CT doctors, a CNN health recognition model is constructed to image reconstruction methods on performance of pulmonary realize personalized diagnosis and treatment of human health. nodules detection algorithm based on deep learning,” Chinese Six subjects are selected to wear wearable measuring devices for Journal of Medical Imaging Technology, vol. 35, no. 12, a long time and physiological parameters are measured as pp. 1775–1779, 2019. required. )e test sample set is input into the deep learning [10] D. Spathis and P. Vlamos, “Diagnosing asthma and chronic model for identification, with a prediction accuracy of over 77%. obstructive pulmonary disease with machine learning,” In the future work, the measurement types of human health Health Informatics Journal, vol. 25, no. 3, pp. 811–827, 2019. parameters will be added, such as ECG and EEG signals, etc. In [11] T. R. Mauldin, M. E. Canby, V. Metsis, A. H. H. Ngu, and addition, the scale of the data set will be expanded, and the C. C. Rivera, “Smartfall: a smartwatch-based fall detection training depth learning model will be updated continuously to system using deep learning,” Sensors, vol. 18, no. 10, pp. 1–19, improve the prediction accuracy. 2018. [12] A. Sathyanarayana, S. Joty, L. Fernandez-Luque et al., “Sleep quality prediction from wearable data using deep learning,” Data Availability JMIR Mhealth and Uhealth, vol. 4, no. 4, p. e125, 2016. [13] X. F. Zhang, X. Li, J. T. Dai et al., “)e design of a hemiplegic )e data used to support the findings of this study are upper limb rehabilitation training system based on surface available from the corresponding author upon request. EMG signals,” Journal of Advanced Mechanical Design, Sys- tems and Manufacturing, vol. 12, no. 1, pp. 1–12, 2018. [14] I. Kaur, K. S. Saini, and J. Khaira, “An effective architecture for Conflicts of Interest smart city using advanced protocols of internet of things,” )e authors declare that they have no conflicts of interest. International Journal of Advanced Science and Technology, vol. 28, no. 19, pp. 803–810, 2019. [15] J. Mathew and R. Jemima Priyadarsini, “Enhancing security in Acknowledgments IoT healthcare services using fog computing,” International Journal of Advanced Science and Technology, vol. 28, no. 17, )e work was supported by the Fundamental Research pp. 444–450, 2019. Funds for Hebei North University (JYT2019004) and [16] S. Pandya, R. Desai, D. Pathak, and K. Kotecha, “IoT based Foundation of the Zhangjiakou Science and Technology intelligent health surveillance & alert system with fault pre- Project (1411073B). diction using machine learning,” International Journal of 10 Journal of Healthcare Engineering Advanced Science and Technology, vol. 28, no. 18, pp. 660–676, [17] W. Chen, T. Kobayashi, S. Ichikawa, Y. Takeuchi, and T. Togawa, “Continuous estimation of systolic blood pressure using the pulse arrival time and intermittent calibration,” Medical, & Biological Engineering & Computing, vol. 38, no. 5, pp. 569–574, 2000. [18] R. Gircys, A. Liutkevicius, E. Kazanavicius, V. Lesauskaite, G. Damuleviciene, and A. Janaviciute, “Photo- plethysmography-based continuous systolic blood pressure estimation method for low processing power wearable de- vices,” Applied Sciences, vol. 9, no. 11, p. 2236, 2019. [19] J. Jilek and T. Fukushima, “Oscillometric blood pressure measurement: the methodology, some observations, and suggestions,” Biomedical Instrumentation & Technology, vol. 39, no. 3, pp. 237–241, 2005. [20] T. Wang, C. Lu, G. Shen, and F. Hong, “Sleep apnea detection from a single-lead ECG signal with automatic feature-ex- traction through a modified LeNet-5 convolutional neural network,” PeerJ, vol. 7, no. 9, p. e7731, 2019. [21] A. Khalaf, E. Sejdic, and M. Akcakaya, “EEG-fTCD hybrid brain-computer interface using template matching and wavelet decomposition,” Journal of Neural Engineering, vol. 16, no. 3, Article ID 036014, 2019. [22] K. Harimoorthy and M. )angavelu, “Multi-disease predic- tion model using improved SVM-radial bias technique in healthcare monitoring system,” Journal of Ambient Intelli- gence and Humanized Computing, vol. 12, no. 3, pp. 3715– 3723, 2020. [23] L. Zhu, I. Reychav, R. McHaney, A. Broda, Y. Tal, and O. Manor, “Combined SNA and LDA methods to understand adverse medical events,” 3e International Journal of Risk & Safety in Medicine, vol. 30, no. 3, pp. 129–153, 2019. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal of Healthcare Engineering Hindawi Publishing Corporation

Smart Healthcare System Based on Cloud-Internet of Things and Deep Learning

Download PDF
10 pages

Loading next page...
 
/lp/hindawi-publishing-corporation/smart-healthcare-system-based-on-cloud-internet-of-things-and-deep-KLU7tFJtby

References (25)

Publisher
Hindawi Publishing Corporation
Copyright
Copyright © 2021 Benzhen Guo 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
10.1155/2021/4109102
Publisher site
See Article on Publisher Site

Abstract

Hindawi Journal of Healthcare Engineering Volume 2021, Article ID 4109102, 10 pages https://doi.org/10.1155/2021/4109102 Research Article Smart Healthcare System Based on Cloud-Internet of Things and Deep Learning Benzhen Guo , Yanli Ma , Jingjing Yang , and Zhihui Wang College of Information Science and Engineering, Hebei North University, 11 Diamond South Road, Zhangjiakou 075000, China Correspondence should be addressed to Yanli Ma; hebeinu_ma@163.com Received 13 April 2021; Revised 31 May 2021; Accepted 20 June 2021; Published 29 June 2021 Academic Editor: Enas Abdulhay Copyright © 2021 Benzhen Guo 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. Introduction. Health monitoring and remote diagnosis can be realized through Smart Healthcare. In view of the existing problems such as simple measurement parameters of wearable devices, huge computing pressure of cloud servers, and lack of individ- ualization of diagnosis, a novel Cloud-Internet of )ings (C-IOT) framework for medical monitoring is put forward. Methods. Smart phones are adopted as gateway devices to achieve data standardization and preprocess to generate health gray-scale map uploaded to the cloud server. )e cloud server realizes the business logic processing and uses the deep learning model to carry out the gray-scale map calculation of health parameters. A deep learning model based on the convolution neural network (CNN) is constructed, in which six volunteers are selected to participate in the experiment, and their health data are marked by private doctors to generate initial data set. Results. Experimental results show the feasibility of the proposed framework. )e test data set is used to test the CNN model after training; the forecast accuracy is over 77.6%. Conclusion. )e CNN model performs well in the recognition of health status. Collectively, this Smart Healthcare System is expected to assist doctors by improving the diagnosis of health status in clinical practice. addition, the importance of feature learning is high- 1.Introduction lighted [7, 8]. )e feature representation of samples in the By the end of 2015, there were about 220 million people original space is transformed into a new feature space by aged 60 or above in China, accounting for 16.1% of the means of feature transformation layer by layer. Deep total population. Among them, 140 million were over 65 learning is widely used in image and voice processing and years old, accounting for 10.5% of the total population. In develops rapidly in the medical and health field [8–10]. this context, healthcare for the elderly has increasingly )e study in [11] proposed a human fall early warning aroused social concern [1]. )e rapid development of algorithm based on RNN and compared it with tradi- wearable devices, IOT, and cloud computing has brought tional machine learning algorithms such as SVM to verify about significant development opportunities and affected its excellent performance. )e study in [12] compared and studied the application of CNN, RNN, LSTM, and every aspect of people’s lives. In the field of healthcare, the adoption of IOT can bring us health monitoring all other deep learning models in the evaluation of sleep the time [2]. With the popularity of smart phones, smart quality. bracelets, and other devices, a variety of sensors can With the increase of the number of IOT devices and monitor health indicators timely and accurately [3–5]. sensors, medical IOT system also highlights more problems Deep learning is an implementation method of machine in the development process, mainly including the following: learning, which is different from the traditional shallow (1) )e measurement parameters of existing wearable models such as artificial neural network. Deep learning health monitoring devices are relatively simple, such usually has a deeper model structure with five or six layers as pedometers and intelligent sphygmomanometers. or even over ten layers or more hidden layers [6]. In )e common pedometer devices can connect with 2 Journal of Healthcare Engineering set. )e newly uploaded data are calculated by a smart phones via Bluetooth and other communi- cation means, and upload data to the cloud server, so depth model to automatically give health assessment results. )e model is a process of dynamic change. it is possible to view data such as exercise amount and sports assessment report through the phone )e private doctor will manually evaluate the user’s APPs. Various forms of wearable health monitoring health status and update the user data set on a regular devices fail to share data due to different manufac- basis as well as the parameter values of the depth turers and communication protocols. model. )e model can realize personalized health assessment. (2) Limited by memory capacity and computing power, wearable devices may upload the collected data to the cloud server through smart phones or home gate- 2. Materials and Methods ways. )e cloud server needs to store and process a 2.1. System Architecture. Figure 1 indicates the architecture large amount of collected data. )e server and of health monitoring system based on C-IOT and deep network have great transmission pressure, which learning proposed in this paper. It is improved from the reduces the real-time processing capacity. traditional three-layer architecture of the IOT, and described (3) At present, many telemedicine monitoring systems from the perspective of system implementation and data based on IOT carry out certain disease early warning flow, including data collection layer, data preprocessing and according to a set of diagnostic schemes, and it is net layer, data processing, and application layer. difficult to develop personalized diagnosis and Data collection layer is responsible for the acquisition of treatment schemes according to individual physio- physiological parameters of health, which is mainly com- logical characteristics and historical parameter posed of a number of Internet-connected or wearable de- changes. vices. Temperature, body fat/weight, blood pressure (4) Intelligent diagnosis methods based on the uploaded parameters, and exercise parameters are measured. In data of health monitoring devices mainly include general, data collection layer devices have small data storage system-based diagnosis methods of experts and in- capacity and low computing capacity [14–16]. In order to telligent diagnosis methods based on sample data. reduce data redundancy, network transmission pressure, Back Propagation (BP) neural network algorithm and power consumption, the data denoising, digital filtering, [13] and Support Vector Machines (SVM) algorithm and power management based on rules can be realized have been applied to the classification of diagnosis locally. High-frequency noise is eliminated by digital low- results. Although deep learning algorithm has been pass filter and band-pass filter. In addition, the special value applied in sleep quality and other fields [12], the in the collected data is removed based on rules. )e data application of intelligent diagnosis algorithm based collection layer device may be connected to smart phones or on deep learning is rarely seen. other mobile intelligent terminals through Bluetooth. In view of the above problems, a healthcare monitoring )e data preprocessing and net layer are mainly com- system based on Cloud-Internet of )ings (C-IOT) and deep posed of smart phones and other intelligent mobile termi- learning is proposed, and the major work includes the nals. Smart phone devices not only serve as network layer following aspects: devices to realize data communication function of medical IOT gateway, but also install application layer APP software (1) A health data acquisition system is designed based on to realize local data preprocessing, parameter display, and C-IOT and the acquisition of parameters such as device control. )e smart phone connects to the data col- human blood pressure, body temperature, body lection layer device through Bluetooth to receive mea- weight/fat, and exercise amount is realized. )e surement information reported by the blood pressure meter, shortcoming that the acquisition of simple physio- weight/fat meter, pedometer, thermometer, and other data logical parameters fails to evaluate and diagnose the acquisition devices, and display the real-time information on user’s health effectively is avoided. the mobile APPs. Preprocessing data are uploaded to the (2) Data preprocessing of each acquisition device is cloud server through Wifi, 4G, and other communication realized locally to eliminate noise interference. By means. Data preprocessing includes data normalization, communicating with the user’s smart phones data dimension transformation, data fusion, and other through Bluetooth, the smart phones may display the functions to generate images of health parameters. In terms collected data in real time, preprocess the data of of data normalization, body weight, body temperature, and each device, and upload them to the cloud server. other measurement parameters are normalized according to )e processing and integration of local data can the grade to the gray value ranging from 0 to 255. In terms of reduce the computing pressure of cloud server and data dimension transformation, original measurement pa- transmission pressure of network. rameters are extended in dimension, such as heart rate, (3) )e private doctor evaluates the user’s health con- systolic pressure and diastolic blood pressure measured by dition according to the data on the cloud server and sphygmomanometer extended to ambulatory pulse pressure establishes the initial data set. )e cloud server uses (APP), mean arterial pressure (MAP), and ambulatory rate- the deep learning algorithm (CNN) to train the user pressure product (ARPP), which are often used to diagnose health evaluation model according to the initial data cardiovascular diseases more effectively. )e application Journal of Healthcare Engineering 3 Elements Function Deep Data AI engine learning storage Data model processing and Private doctor training Intelligent Web application report Deep layer learning Model Server/database model update activition Users Router Data Data Data normalization transmission preprocessing Smartphone and net Data dimension Health image layer enhancement generation PAD Blue tooth Blood pressure Pedometer Data Data detection filtration Data collection layer Weight/fat Temperature Rule-based Power denoising management Figure 1: System architecture. software of mobile phones can also fuse the physiological trained deep learning model is used to solve the health pa- rameter image. With the increase of user data, private doctors parameter measurement data after normalization of gray value and dimension expansion into the image of health can annotate the data, enrich the personal health data set, and retrain and update the deep learning model. parameters, upload them to the cloud server in the form of two-dimensional image, and use the deep learning model to solve the health parameter image to give the health index 2.2. Measurement of Blood Pressure. For blood pressure report. )e users may receive and display the health report acquisition nodes, the oscillometric method is used to issued by the cloud server through smart phones. Making measure blood pressure [17], and the pressure value of the full use of the computing power of intelligent equipment for pressure sensor is filtered by low-pass filter and band-pass data preprocessing can effectively reduce the transmission filter to obtain the static pressure value and the dynamic pressure of database and network. With the rapid im- pressure value. In the measurement process, the dynamic provement of the computing power of intelligent terminal pressure value amplitude increases gradually and then de- creases. When the dynamic pressure value amplitude is equipment, more and more data processing functions will be completed directly in the intelligent terminals. multiplied by the normalized coefficients (Ks and Kd), the )e data processing and application layer mainly includes dynamic pressure value amplitude corresponding to the database and distributed server, web server, and deep learning systolic pressure and the diastolic pressure can be obtained, model engine. )e database stores and manages users’ per- respectively, and the human blood pressure value can be sonal information/health monitoring data, personal doctor obtained by reverse check, as shown in Figure 2. According information, equipment information, etc., the distributed to the conclusions of Mauro’s mathematical model, the server realizes various business processing logic, and the web normalized coefficient Ks was 0.46–0.64, and the normalized server provides users and personal doctors with friendly web coefficient Kd was 0.43–0.73 [18]. In this paper, the am- interface for background operation. )e deep learning model plitude coefficients commonly used in clinical medicine are engine is used to train the deep learning model, and the Ks � 0.48 and Kd � 0.58 [19]. 4 Journal of Healthcare Engineering fitting: a set of sample data (x , y )(i � 1, 2, 3 . . . N) can be i i At described by Gaussian functions as shown in the following equation: x − x 􏼁 i max (1) y � y ∗ exp􏼢− 􏼣. i max Kd = Ad/A max Ks = As/A max Take the logarithm of both sides of this equation: Ps Pc Pd Pt Figure 2: Blood pressure measurement method. In order to obtain the optimal pulse oscillation ampli- tude envelope, Gaussian fitting method is used for data 2 2 2 x − x 􏼁 x 2x x x i max max i max i (2) ln y � ln y − � 􏼠ln y − 􏼡 + − . i max max s s s s Assume preprocessing shall be conducted through the application software installed in the smart phone, which mainly includes ln y � z , i i dimensional transformation of data, data standardization, and generation of two-dimensional gray image for human max health. ln y − � b , max 0 In order to optimize the input vector, the measurement (3) data uploaded by monitoring equipment need to be 2x max � b , transformed into a certain dimension. )e measured data including heart rate (HR), systolic blood pressure (SP), and diastolic blood pressure (DP) are converted into three pa- − � b . rameters, namely, mean arterial pressure (MAP), ambula- tory pulse pressure (APP), and ambulatory rate-pressure Taking all sample data into consideration, equation (2) is product (ARPP). converted into a matrix as follows: APP � F1(SP, DP) � SP − DP, z 1 x x 1 1 ⎡ ⎢ ⎤ ⎥ ⎢ ⎥ ⎡ ⎢ ⎤ ⎥ ⎢ ⎥ b ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ 0 ⎢ ⎥ ⎢ ⎥ DP + (SP − DP) ⎢ ⎥ ⎢ 2 ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ z ⎥ ⎢ ⎥⎡ ⎢ ⎤ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ 1 x x ⎥ (7) ⎢ 2 ⎥ ⎢ ⎢ ⎥ MAP � F2(SP, DP) � , ⎢ ⎥ ⎢ 2 2 ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ � ⎢ ⎥⎢ b ⎥. (4) ⎢ ⎥ ⎢ ⎥⎢ ⎥ 3 ⎢ ⎥ ⎢ ⎥⎢ 1 ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎣ ⎦ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢⋮ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢⋮ ⋮ ⋮ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎣ ⎦ ⎢ ⎥ ⎣ ⎦ 2 ARPP � F3(HR, SP) � HR∗ SP. 1 x x n n Step number, motion distance, fast motion time, fast It is simplified as motion distance, length of sleeping, and length of awakening Z � XB. (5) are obtained according to the step number and time mea- sured by the pedometer after dimension expansion. Data According to the principle of least squares, the gener- standardization converts the data of each index into the gray alized solution of least squares for matrix B is value ranged from 0 to 255 according to the scaling, so as to − 1 T T facilitate the synthesis of health status matrix for uploading (6) B � 􏼐X X􏼑 X Z. to the cloud server for processing. )e corresponding range relationship between the measured value of each index and )en, the estimated parameters (b , b , b ) are 0 1 2 the standardized value is shown in Table 1. substituted into equation (1) to obtain the fitted Gaussian In this paper, the deep learning method is adopted to function. Both the measuring speed and accuracy are im- evaluate the human health status. )e input data of the deep proved by using Gaussian fitting function. learning model are the measured values after the stan- dardization of each physiological parameter of human body in a day. In order to better serve as the input of CNN, the 2.3. Data Preprocessing. After the human physiological standardized data need to be processed in two dimensions to parameter monitoring equipment transmits the measure- generate the health status matrix. )e health status matrix is ment results to a smart phone via Bluetooth, data Journal of Healthcare Engineering 5 Table 1: Data standardization. Index Range of measurement value Range of standard value Weight 0–150 kg 0–255 Fat 0%–50% 0–255 SP 0–250 mmHg 0–255 DP 0–250 mmHg 0–255 HR 0–200 bpm 0–255 APP 0–200 mmHg 0–255 MAP 0–300 mmHg 0–255 ARPP 0–50000 0–255 Step num 0–30000 0–255 Distance 0–20 km 0–255 Fast motion time 0–24 h 0–255 Fast motion distance 0–20 km 0–255 Length of sleeping 0–24 h 0–255 Length of awakening 0–24 h 0–255 Body temperature 30–45 C 0–255 organized into 2D images. As required, the subjects shall 3 and 9 filters, respectively, and a scaling of 2 for the max- have their blood pressure and body temperature measured pooling layers. )e output layer outputs the health as- once every morning and every evening, have their weight/fat sessment results, including health, sub-health, and measured once every day, and wear a pedometer 24 hours a illness. day. On the smart phone side, 36-pixel 2D images of human health are generated every day, as shown in Table 2. Figure 3 3. Results and Discussion is an example 2D image of health status matrix according to 3.1. Measurement and Test of Blood Pressure. )e method of Table 2. using medical blood pressure meter and blood pressure measurement node at the same time for the same subject 2.4.CollectionofDataSet. )e training of deep learning model is used for comparative verification. A health monitor shall be supported by a certain amount of labeled data set. In this (PM-900S, Biocare Technology Co. Ltd., Shenzhen, study, the gray-scale chart of health monitoring parameters in China) is selected as the comparison device. )e verifi- one day is obtained by combining measurement and con- cation results are shown in Table 3, and the measurement struction. Six volunteers are selected, including 3 males and 3 results indicate that the relative error is within the range females (including 2 adolescents, 2 middle-aged, and 2 elderly). of 6%. Figure 5(a) is the photo of blood pressure mea- )e volunteers are required to test the blood pressure each surement node device. morning and evening, the body temperature once, and the Also, the method of wearing commercial pedometer weight/fat once. )ey are also required to wear a pedometer and step monitoring node is used for comparative ver- device for 24 hours and upload data to the cloud server at 6 ification. A sports bracelet (Honor 3, HUAWEI Tech- o’clock each morning through the phone APP. )en, private nology Co. LTD, Shenzhen, China) is selected as the doctors will grade their health status according to the uploaded comparison experiment device. )e test subjects wear data and user information such as age and gender and then sports bracelets and step monitoring nodes at the same divide them into three types: health, sub-health, and illness. )e time. )e measured data of the previous day is read after initial data set is generated by continuously tracking and an- the subjects get up at 6 o’clock in the morning. )e notating the user’s data for 100 days. In addition, the initial data verification results are shown in Table 4, and the mea- set of each user constructed and annotated by private doctors surement results indicate that the relative error is within according to the user’s historical data information is 3,000, and the range of 9%. Figure 5(b) is the photo of step moni- these three types account for 1/3. 80% of the extended initial toring node. data set is selected as the training sample to train the user’s personalized deep learning model, and 20% is used as the test sample to test the model’s performance. 3.2. Evaluation of Deep Learning Algorithm and Model. For each subject, the accuracy of average recognition of 2.5. Deep Learning Model. Deep learning model is con- the three health states is studied. Apart from that, recall structed and trained in the cloud server, and CNN is the and precision and F1_score are used for model evaluation most commonly used deep learning model, which has in respect of each category [21]. )e data of test set in data been widely used in the field of image processing. )e set are used as model input. TP denotes number of true classic LeNet-5 [20] CNN model is adopted and modified positive (labeled correctly). FP denotes number of false to simplify a pooling layer. )e CNN model constructed positive (other activity labeled as the sub-health and is shown in Figure 4, which contains two convolution illness). Furthermore, TN denotes number of true neg- layers, a pooling layer, and a full connection layer. )e atives (correct rejection), and FN denotes number of false CNNs have 2 × 2 kernels for the convolutional layers with negatives (missed detections). 6 Journal of Healthcare Engineering Table 2: Human health map. 1 2 3 4 5 6 1 SP1 DP1 HR1 APP1 MAP1 ARPP1 2 SP2 DP2 HR2 APP2 MAP2 ARPP2 3 Weight Fat 4 Step num Distance Fast motion time Fast motion distance Length of sleeping Length of awakening 5 Body temperature1 6 Body temperature2 0 12345 Figure 3: 2D images of human health. C1: 4 × 4@9 S1: 2 × 2@9 C1: 5 × 5@3 2 150 0 12345 Subsampling layer Max pooling: 2 × 2 Convolution layer Kernel size: 2 × 2 Convolution layer Full connection Somax Filter number: 3 Kernel size: 2 × 2 layer Stride length: 1 Filter number: 9 Stride length: 1 Figure 4: CNN model architecture. (S4) and the lowest is 68.5% (S3). )e standard deviation TP + TN Accuracy � , of recognition accuracy is all within 15 and the maximum TP + FP + FN + TN is 14.94 (S5). For 6 subjects, the average accuracy of CNN TP model is 77.61%. recall � , Figure 7 indicates the confusion matrices of 6 subjects TP + FN (8) acquired by CNN model. It can be revealed that the rec- TP ognition performance of the model for health categories is precision � , TP + FP obviously superior to the other two categories. In 200 samples, the highest recognition accuracy is 182 (S6). )e 2precision∗ recall F1 score � . capability to recognize sub-health categories is relatively precision + recall poor, and S5 can only recognize 110 out of 200 samples Figure 6 exhibits the accuracy of the average classi- correctly. fication of each subject. Error bars display the standard Figure 8 shows the precision, recall, and F1-score of 6 deviation of the recognition accuracy of the three health subjects in three categories acquired by CNN model. It categories for each subject. As can be seen from the figure, can be seen that CNN model boasts higher precision for the results of different subjects are of obvious differences. disease recognition than other two categories, except S2. )e highest recognition accuracy of CNN model is 84.2% Nevertheless, the value of recall is higher than that of Journal of Healthcare Engineering 7 Table 3: Blood pressure measurement verification results. SP (mmHg) (PM-900S) SP (mmHg) (test node) Error (%) DP (mmHg) (PM-900S) DP (mmHg) (test node) Error (%) 1 120 123 2.5 83 86 3.6 2 107 112 4.7 78 82 5.1 3 124 130 4.8 89 86 3.4 4 109 105 3.7 78 81 3.8 5 140 145 3.6 95 99 4.2 (a) (b) Figure 5: (a) Blood pressure measurement node device. (b) Step monitoring node device. Table 4: Step monitoring node verification results. Step num (honor3) Step num (test node) Error (%) Amount of sleep (h) (honor3) Amount of sleep (h) (test node) Error (%) 1 5600 5743 2.6 6.5 6.3 3.1 2 8212 7920 3.6 7.4 7.0 5.4 3 15401 16280 5.7 8.2 7.7 6.1 4 7231 7102 1.8 7.1 6.6 7.0 5 11005 10098 8.2 9.4 9.0 4.3 S1 S2 S3 S4 S5 S6 AVG CNN Figure 6: Classification accuracy of each subject. Each bar and the corresponding error bar show the average classification accuracy with standard deviation of three health patterns. Accuracy (%) 8 Journal of Healthcare Engineering Health Subhealth Illness Health Subhealth Illness Health Subhealth Illness 160 140 Health 171 18 11 Health 163 20 17 Health 150 34 16 140 120 Subhealth 30 152 18 Subhealth 22 140 38 Subhealth 46 120 34 Illness 922 169 Illness 10 12 178 Illness 25 34 141 20 20 (a) (b) (c) Health Subhealth Illness Health Subhealth Illness Health Subhealth Illness Health 180 12 8 Health 167 21 12 140 Health 182 11 7 100 100 Subhealth 28 153 19 Subhealth 49 110 41 Subhealth 32 145 23 60 60 40 40 40 Illness 12 16 172 Illness 21 25 154 Illness 33 20 147 20 20 (d) (e) (f) Figure 7: Confusion matrices acquired by the CNN model. (a–f) )e confusion matrix of S1–S6. 100 100 100 95 95 95 90 90 90 85 85 85 80 80 80 75 75 75 70 70 70 65 65 65 60 60 60 Health Subhealth Illness Health Subhealth Illness Health Subhealth Illness Precision (%) Precision (%) Precision (%) Recall (%) Recall (%) Recall (%) F1_score F1_score F1_score (a) (b) (c) 100 100 100 95 95 90 90 85 85 80 80 75 75 70 70 65 65 60 50 60 Health Subhealth Illness Health Subhealth Illness Health Subhealth Illness Precision (%) Precision (%) Precision (%) Recall (%) Recall (%) Recall (%) F1_score F1_score F1_score (d) (e) (f) Figure 8: Precision, recall, and F1-score obtained by the CNN model. (a–f) )e result curves of S1–S6. precision except S2. F1_score is a comprehensive index )e current research suggests that it is feasible to recognize and intelligently diagnose the preprocessed reflecting the performance of the model. By observing the F1_score curve of 6 subjects, the recognition performance data of the underlying devices of the medical Internet of of the model for sub-health is worse than the other two )ings with CNN model. In comparison with traditional categories, with a minimum of 61.8% (S5). machine learning methods such as SVM [22] and LDA Journal of Healthcare Engineering 9 [23], the depth learning method for intelligent diagnosis References does not require manual selection of data feature values. [1] S. Din and A. Paul, “Retracted: smart health monitoring and )e depth model can automatically extract features and management system: toward autonomous wearable sensing perform high-level abstraction. For different subjects, the for Internet of )ings using big data analytics,” Future optimal parameters of the model can be automatically Generation Computer Systems, vol. 91, pp. 611–619, 2019. acquired through training, which is more flexible and [2] W. Shi, J. Cao, Q. Zhang, Y. Li, and L. Xu, “Edge computing: robust compared with traditional machine learning vision and challenges,” IEEE Internet of 3ings Journal, vol. 3, methods. )e intelligent health management system ar- no. 5, pp. 637–646, 2016. chitecture based on deep learning proposed in the current [3] B. Kwolek and M. Kepski, “Human fall detection on em- research can make the data set continuously grow under bedded platform using depth maps and wireless accelerom- the condition of user data accumulation, and cooperate eter,” Computer Methods and Programs in Biomedicine, with doctors’ manual annotation, and can optimize vol. 117, no. 3, pp. 489–501, 2014. model training from time to time to achieve better rec- [4] Y. Kim and T. Moon, “Human detection and activity clas- sification based on micro-doppler signatures using deep ognition and diagnosis effects. Pretreatment of health convolutional neural networks,” IEEE Geoscience and Remote data at the bottom layer of Internet of )ings equipment Sensing Letters, vol. 13, no. 1, pp. 8–12, 2016. enhances the robustness of data and can tremendously [5] M. Z. Uddin, “A wearable sensor-based activity prediction lower the computing pressure of servers. system to facilitate edge computing in smart healthcare For each subject, the accuracy of average recognition of system,” Journal of Parallel and Distributed Computing, the three health states is studied. Apart from that, recall and vol. 123, pp. 46–53, 2019. precision and F1_score are used for model evaluation in [6] L. I. U. Dong, L. I. Yue, L. I. N. Jianping, L. I. Houqiang, and respect of each category. W. U. Feng, “Deep learning-based video coding: a review and a case study,” ACM Computing Surveys, vol. 53, no. 1, pp. 1–35, 4.Conclusions 2020. [7] M. Ghorbani, M. Bahaghighat, Q. Xin, and F. Ozen, “Con- In view of the existing problems in the existing IOT medical vLSTMConv network: a deep learning approach for sentiment system, a new IOT architecture for medical monitoring is analysis in cloud computing,” Journal of Cloud Computing, proposed in this paper, in which smart phones are used as vol. 9, no. 1, pp. 1–13, 2020. [8] L. Zheng, O. Wang, S. Hao et al., “Development of an early- gateway devices to realize data preprocessing of measurement warning system for high-risk patients for suicide attempt node devices, thus greatly reducing the computing pressure of using deep learning and electronic health records,” Transla- cloud servers and transmission pressure of network. Based on tional Psychiatry, vol. 10, no. 1, pp. 1–11, 2020. the data sets generated from such data annotated by private [9] Z. Liu, Y. Fu, Y. Zhang, F. Peng, and Z. Zhang, “Effect of CT doctors, a CNN health recognition model is constructed to image reconstruction methods on performance of pulmonary realize personalized diagnosis and treatment of human health. nodules detection algorithm based on deep learning,” Chinese Six subjects are selected to wear wearable measuring devices for Journal of Medical Imaging Technology, vol. 35, no. 12, a long time and physiological parameters are measured as pp. 1775–1779, 2019. required. )e test sample set is input into the deep learning [10] D. Spathis and P. Vlamos, “Diagnosing asthma and chronic model for identification, with a prediction accuracy of over 77%. obstructive pulmonary disease with machine learning,” In the future work, the measurement types of human health Health Informatics Journal, vol. 25, no. 3, pp. 811–827, 2019. parameters will be added, such as ECG and EEG signals, etc. In [11] T. R. Mauldin, M. E. Canby, V. Metsis, A. H. H. Ngu, and addition, the scale of the data set will be expanded, and the C. C. Rivera, “Smartfall: a smartwatch-based fall detection training depth learning model will be updated continuously to system using deep learning,” Sensors, vol. 18, no. 10, pp. 1–19, improve the prediction accuracy. 2018. [12] A. Sathyanarayana, S. Joty, L. Fernandez-Luque et al., “Sleep quality prediction from wearable data using deep learning,” Data Availability JMIR Mhealth and Uhealth, vol. 4, no. 4, p. e125, 2016. [13] X. F. Zhang, X. Li, J. T. Dai et al., “)e design of a hemiplegic )e data used to support the findings of this study are upper limb rehabilitation training system based on surface available from the corresponding author upon request. EMG signals,” Journal of Advanced Mechanical Design, Sys- tems and Manufacturing, vol. 12, no. 1, pp. 1–12, 2018. [14] I. Kaur, K. S. Saini, and J. Khaira, “An effective architecture for Conflicts of Interest smart city using advanced protocols of internet of things,” )e authors declare that they have no conflicts of interest. International Journal of Advanced Science and Technology, vol. 28, no. 19, pp. 803–810, 2019. [15] J. Mathew and R. Jemima Priyadarsini, “Enhancing security in Acknowledgments IoT healthcare services using fog computing,” International Journal of Advanced Science and Technology, vol. 28, no. 17, )e work was supported by the Fundamental Research pp. 444–450, 2019. Funds for Hebei North University (JYT2019004) and [16] S. Pandya, R. Desai, D. Pathak, and K. Kotecha, “IoT based Foundation of the Zhangjiakou Science and Technology intelligent health surveillance & alert system with fault pre- Project (1411073B). diction using machine learning,” International Journal of 10 Journal of Healthcare Engineering Advanced Science and Technology, vol. 28, no. 18, pp. 660–676, [17] W. Chen, T. Kobayashi, S. Ichikawa, Y. Takeuchi, and T. Togawa, “Continuous estimation of systolic blood pressure using the pulse arrival time and intermittent calibration,” Medical, & Biological Engineering & Computing, vol. 38, no. 5, pp. 569–574, 2000. [18] R. Gircys, A. Liutkevicius, E. Kazanavicius, V. Lesauskaite, G. Damuleviciene, and A. Janaviciute, “Photo- plethysmography-based continuous systolic blood pressure estimation method for low processing power wearable de- vices,” Applied Sciences, vol. 9, no. 11, p. 2236, 2019. [19] J. Jilek and T. Fukushima, “Oscillometric blood pressure measurement: the methodology, some observations, and suggestions,” Biomedical Instrumentation & Technology, vol. 39, no. 3, pp. 237–241, 2005. [20] T. Wang, C. Lu, G. Shen, and F. Hong, “Sleep apnea detection from a single-lead ECG signal with automatic feature-ex- traction through a modified LeNet-5 convolutional neural network,” PeerJ, vol. 7, no. 9, p. e7731, 2019. [21] A. Khalaf, E. Sejdic, and M. Akcakaya, “EEG-fTCD hybrid brain-computer interface using template matching and wavelet decomposition,” Journal of Neural Engineering, vol. 16, no. 3, Article ID 036014, 2019. [22] K. Harimoorthy and M. )angavelu, “Multi-disease predic- tion model using improved SVM-radial bias technique in healthcare monitoring system,” Journal of Ambient Intelli- gence and Humanized Computing, vol. 12, no. 3, pp. 3715– 3723, 2020. [23] L. Zhu, I. Reychav, R. McHaney, A. Broda, Y. Tal, and O. Manor, “Combined SNA and LDA methods to understand adverse medical events,” 3e International Journal of Risk & Safety in Medicine, vol. 30, no. 3, pp. 129–153, 2019.

Journal

Journal of Healthcare Engineering – Hindawi Publishing Corporation

Published: Jun 29, 2021

There are no references for this article.