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

Learn More →

AORTA software system for evaluating individual predisposition to atherosclerosis on the basis of genetic and phenotypic markers

AORTA software system for evaluating individual predisposition to atherosclerosis on the basis of... This article deals with the AORTA software system providing support for research activities to find molecular basis for further assessment of individual predisposition to atherosclerosis. These studies are aimed at finding a relationship between somatic mutations of the mitochondrial genome in the aortic wall cells and the extent of atherosclerotic lesions of the aorta. A morphologist selects these areas on an aortic tissue sample and describes them, so that within each area, deviation of the quantitative indicator of atherosclerosis severity (phenotypic marker) from the area average should be sufficiently small. Next, the frequency and severity indicators of somatic mutations of the mitochondrial genome (genetic markers) are measured for each area and then entered into the AORTA system. Keywords: aortic wall; atherosclerosis; genetic and phenotypic markers; image processing; somatic mutations. DOI 10.1515/bams-2014-0023 Received December 21, 2014; accepted February 2, 2015 Introduction Currently, automation of medical and biomedical research is an important scientific and technical task. Identification of patients' predisposition to various diseases is of special importance in this research. This article deals with the development of AORTA system providing support for research carried out in the Russian Cardiology Research and Production Complex and the Institute of General Pathology and Pathophysiology of the Russian Academy *Corresponding author: Anatoly Karpenko, Computer-Aided Design, Bauman Moscow State Technical University, 2-ya Baumanskaya ul. 5, Moscow 105005, Russian Federation, E-mail: apkarpenko@mail.ru Valery Kharlamov: MSTU n.a. Bauman, Moscow, Russian Federation Igor Sobenin: Russian Cardiology Research and Production Complex, 3-ya Cherepkovskaya 15-a, 121552, Moscow, Russian Federation; and Institute of General Pathology and Patophysiology, Baltiyskaya ul., 5, 125315, Moscow, Russian Federation of Medical Sciences. These studies are aimed at finding a relationship between somatic mutations of the mitochondrial genome in the aortic wall cells and the extent of atherosclerotic lesions of the aorta [1]. Mitochondrial mutations were identified in various pathologies, such as coronary stenosis, some forms of type 2 diabetes mellitus and deafness, atherosclerosis, myocardial infarction, and cardiomyopathy. The mitochondrial genome is notable for its pronounced instability; therefore, frequent somatic mutations arise during the lifetime of an individual. Penetrance and expression of such mutations depend mainly on the genotype and the level of heteroplasmy. The latter represents a proportion of the number of the mitochondrial DeoxyriboNucleic Acid (mtDNA) copies bearing mutation (single nucleotide substitution, insertion, or deletion) at some position of the total number of mtDNA copies. Therefore, when studying the association of mitochondrial mutations with human disease, not only qualitative assessment (the presence/ absence of mutation) but also quantitative assessment of the mutant allele of the mitochondrial genome (the level of heteroplasmy) is necessary. To determine the critical level of heteroplasmy of mitochondrial mutations associated with formation and development of pathologies in humans, a method for direct quantitative assessment of the mitochondrial genome was developed [1]. As initial data for studying the role of mtDNA polymorphisms in the formation of atherosclerotic lesions of human arteries, the AORTA system uses quantitative severity of atherosclerotic lesions and mitochondrial mutations, as well as geometric shapes of the aorta areas under study. A morphologist selects these areas on an aortic tissue sample and describes them as unaffected regions, initial lesions, fatty streaks, lipofibrous, or fibrous atherosclerotic plaques, so that within each area, deviation of the quantitative indicator of atherosclerosis severity (phenotypic marker) from the area average should be sufficiently small. Next, total DNA is extracted from each morphologically identified aortic intimal sample, and the level of heteroplasmy for certain mtDNA mutations is measured in a quantitative manner. Therefore, the frequency and severity indicators of somatic mutations of the mitochondrial 26Karpenko et al.: Image processing genome (genetic markers) are measured for each area and then entered into the AORTA system. Analogues of the AORTA system are unknown to the authors. together with the results of their processing, and also, acting in the automatic mode, finds the boundaries of areas that have homogeneous lesions. The users should be able to associate the extensible set of properties with each of the areas: phenotypic markers, genetic markers, etc. Image processing must be carried out using an extensible set of processing modules. The sequence of using these modules must be user-defined. Problem statement The general view of the samples of human aorta derived from autopsy study is shown in Figure 1. In this figure, we can see selected areas within each of which the phenotypic marker (namely, the extent of atherosclerotic lesion) deviations from the area average are sufficiently small. The areas having a greater extent of atherosclerotic lesions appear saturated red. The areas are numbered consecutively, and regardless of the extent of atherosclerotic lesion, a unique ID number is provided for each sample. In its database, the AORTA system stores the images of aortic fragments that have atherosclerotic lesions, Methodology of solving the problem The proposed methodology of solving the problem consists of the following main stages: 1. image segmentation. 2. connected-component search and labeling. 3. graphing region adjacency of the labeled image. 4. adjacency graph partitioning. Image segmentation Image segmentation can be defined as a process of assigning labels to each pixel in the image so that sets of pixels with the same labels have common visual characteristics. Image segmentation results in a set of segments covering the image or a set of contours of these segments [2­4]. The following segmentation methods are best known: cluster analysis methods, including the iterative k-means algorithm [3]; histogram methods, which are non-iterative clustering methods [3]; methods of separation and fusion of areas [5]; area growth methods [6]; edge detection methods [5]; segmentation methods based on the morphological watershed algorithm, using an image as a three-dimensional surface defined by two spatial coordinates and a brightness level as surface elevation (relief) [5]; methods based on graph theory, for example, methods of normalized cut [7] and segmentation by weighted aggregation [8]; neural network methods [9, 10]. Our experiments with the methods of image segmentation showed that none of them could perform, with satisfactory accuracy, either automatic or automated selection of boundaries of the areas of aortic intima differing by the extent of atherosclerotic lesion under study. Therefore, it was decided that the user marks out the image manually and the AORTA system reads the marked image and automatically selects area boundaries. To solve the latter problem, we use global threshold binarization in one-dimensional attribute space by manually adjusting Figure 1:Photos of the aorta tissue samples (the areas that have homogeneous atherosclerotic lesions are selected). Karpenko et al.: Image processing27 the threshold. As an identifier, pixel brightness is used for monochrome images, and a scalar function of pixel color components is used for color images. Selection (marking) of connected components in an image B(r, c){0; 1} denotes the value of the pixel located at the intersection of the row r and the column c of the pixel array formed at the previous stage. We introduce a 4-connected N4(r, c) and an 8-connected N8(r, c) neighbors of this pixel. We assume that pixel (r, c) is the N4-neighbor to pixel (r, c), if (r, c)N4(r, c). Similarly, we define the N8-neighbor of pixel (r, c) to pixel (r, c). Pixel (r, c) is called the N4-connective to pixel (r, c) by the value v{0; 1} if there exists such a sequence of pixels (ri, ci), i=0, 1, 2, ..., n, (r0, c0)=(r, c), (rn, cn)=(r, c) that pixel (ri­1, ci­1)N4(ri, ci) and B(ri, ci)=v for all i[1:n]. The given sequence of pixels (ri, ci), i=0, 1, 2, ..., n, forms the connected N4-path from pixel (r, c) to pixel (r, c). Similarly, we define the N8-path from pixel (r, c) to pixel (r, c). The connected component of the pixel array B with the value v is formed by a set of C pixels having the value v, each pair of which is connected by the value v [11]. In the given symbols, labeling the connected components of the binary array B is as follows: select all the connected components in the given array and form a labeled image in which each pixel in the image is assigned a label (unique identifier) of the connected component to which the given pixel belongs. Allocation algorithms of connected components are divided into the following five classes [11]: multipass algorithms, two-pass algorithms, single-pass algorithms, algorithms using hierarchical structures of image representation, and parallel algorithms. Multipass algorithms use multiple scanning of images [12, 13]. Two-pass algorithms are classic allocation algorithms of connected components [14]. Single-pass algorithms form a class of the simplest connected components labeling algorithms. A single-pass algorithm scheme is as follows: 1. Find an unlabeled pixel with the value v=1 and assign it a new label. 2. Call the search procedure for all unprocessed neighbors of the given pixel with the same value v=1. 3. For each of the neighbors found, produce a recursive call of the search procedure and so on. The algorithm's search procedure can be constructed basing on the turn-and-stack data structures. In the first case, the algorithm is called breadth-first search algorithm, and in the second case, it is called depth-first search (DFS) algorithm. One of the most efficient singlepass algorithms is the contour tracing algorithm proposed by Chang et al. [15]. There is a highly effective image representation method called run-length encoding [16­18]. Due to high effectiveness and ease of implementation, the single-pass DFS algorithm having complexity O(n), where n is the number of pixels in the image, is used in the AORTA system. Graphing region adjacency of the labeled image Two regions of the image are called adjacent if some pixel in one of the regions is adjacent to some other pixel in another region. Information on related regions of the image is usually represented as region adjacency graph (RAG), where each node corresponds to a region of the image and the edges connect pairs of nodes corresponding to the adjacent regions. The AORTA system uses the algorithm for graphing region adjacency presented in the paper [3]. The algorithm processes the image by taking two lines at a time, showing horizontal and vertical adjacency, and if the N8-neighbor is considered, it also shows diagonal adjacency. When a new region adjacency is found, a new edge is added to RAG. Graph partitioning Suppose G=V, E is a given undirected weighted graph, where V is a vertex set and E is a set of edges. Graph partitioning is as follows: it is necessary to divide set V by k of the subsets Vi, i[1: k] so that any two subgraphs Vi, Vj, ij have no common vertices and that each of the vertices of the graph G belongs to one of the subgraphs. Graph partitioning is usually subordinate to the extremalization of some partitioning quality criterion. Therefore, we will talk about an optimal graph partitioning. This problem is known to be NP-hard. On this basis, in computation, we usually use the following heuristic methods of problem solving. The Kernighan-Lin algorithm is used to improve initial partitioning of the graph by sharing vertices among subgraphs so as to reduce the total number of cut edges [18]. Levelized nested dissection (LND) is used for finding a bisection of the graph that provides equality of the numbers of nodes in resulting subgraphs [19]. The spectral graph bisection algorithm uses two partitioning optimality criteria ­ minimal cut edges number and minimal difference between vertex numbers in different 28Karpenko et al.: Image processing subgraphs. The algorithm is based on the analysis of the eigenvector corresponding to the second largest eigenvalue of the original graph's Laplacian matrix [3]. Multilevel k-way partitioning [20] provides graph partitioning in three stages: coarsening of the original graph, partitioning the coarsened graph into k subgraphs, and recovery of the original graph. The latter algorithm is used for graph partitioning in the AORTA system. The database interface program includes (Figure 3) a data editor and a data processing subsystem. The editor allows the user to load data into the database, perform visualization of the data that are already in the database, modify the data, and submit the modification results to the database. The processing subsystem is a set of independent plug-ins for data processing as well as a shell that provides sharing of the modules. The processing subsystem also implements the user interface. We use MySQL as a database management system (DBMS). The main reason for this choice is to support a large number of MySQL table types, for example, MyISAM tables for full text search and InnoDB tables for transactions at the level of individual records. The application was developed on the.NET Framework 3.5 platform using the C# programming language. Microsoft Visual Studio 2008 was used as an integrated development environment; it has advanced graphical user interface programming aids and provides ample opportunities for project management and debugging. MySQL Connector/NET library was used to interact with the MySQL DBMS. This library implements all the necessary interfaces of ADO.NET, which is part of the.NET Framework providing the programmer with uniform access to different data sources. The Windows Forms library was used to create the graphical user interface. The AORTA bundled software Structure of the AORTA bundled software The structure of the AORTA bundled software is shown in Figure 2. The bundled software consists of a relational database and its software interface, that is, the front end of the system. < > < > : Workstation < > : Database server < > : .NET Framework < > : MySQL Server < > : AORTA.exe : Database schema FormView controls TCP/IP 1 Figure 2:Structure of the AORTA system. FormView controls include "login", the main "AORTA" view, "data editor", "selecting image boundaries", "color < > Program system < > Data editor < > Set of plug-ins IDataStreamProcessor < > Data processing subsystem < > < > MySQL Connector/NET < > .NET Framework < > : AORTA.exe Figure 3:Diagram of components. Karpenko et al.: Image processing29 gradient assignment", "data processing manager", "data filtering", "setting constraints", and "general settings". Login This FormView control is used to set the username and password for further authentication; it also contains fields for setting network address and a port to connect to a MySQL Server Instance (these fields are available only when the application is run under administrator account). sample regions on which relative character values of the regions are indicated by means of color gamut. To configure the imaging parameters, a tool for setting a linear gradient is included in the graphics editor. The left side of the data editor FormView control contains tables to display text and numeric data. The right side displays a graphical image map of aortic regions selected for studying a sample of the aorta. At the bottom, there is a stack for displaying auxiliary data. Selecting image boundaries The FormView control (Figure 5) allows the user to select an area of the aorta. The user's work with this form begins with loading the image of the aorta sample from the file. Next, the user can manually select region boundaries (for example, using a mouse) or use built-in image segmentation. Manual selection of boundaries includes the brush tool and the eraser tool, the parameters of which can be modified using interface elements located in the toolbar on the right side of the form. As mentioned above, the built-in segmentation tools support threshold binarization of the loaded image with manual setting of intensity threshold. To help the user select the correct threshold, a pixel intensity histogram of the image is displayed in the toolbar. The result of setting image boundaries is a binary image of the aortic area under study. Pressing the refresh button implements image splitting ­ that is, it implements automatic search and labeling of the connected The main application window If authentication is successful, the user menu is available in the main application window. Accessibility (activation capability) to menu items is defined by the active user privileges. Data editor This FormView control (Figure 4) allows the user to input data into the database, editing, and visualization. To set geometric shapes of the aortic areas, a graphics editor allowing to set positions of region boundaries manually or use image segmentation tools is included in the data editor. In the latter case, threshold image binarization (item 1) is implemented and intensity threshold is set manually. Data visualization includes a map of aorta Figure 4:Data editor. 30Karpenko et al.: Image processing one can enable and disable display of each of the layers as well as image zoom. Color gradient assignment The color gradient assignment FormView allows the user to select one of the predefined color gradients or specify a custom gradient. Piecewise linear color gradient is presented in the form of the vector-function G(n)=(gR(n), gG(n), gB(n)), whose values are RGB color components. The argument of this function is a normalized color value nR, n[0; 1]. The value of the gradient is determined basing on piecewise linear interpolation of color components in userdefined color stops for which corresponding values n (in percentage terms) are defined. The user can change the number of color stops, their location, and color. Figure 5:Selecting image boundaries. Data processing manager This FormView (Figure 6) provides control tools for data processing sequence (pipeline). The left side of the form displays a list of available processing modules (processors), their characteristics, and description. Processors can have their own user interfaces; so, in general, the pipeline presupposes interaction with the user. The right side of the form allows the user to create new pipelines, select, and edit previously formed pipelines. This part of the form allows to run the selected pipeline in a singlestep or nonstop mode. components of the image, removing insufficiently large areas, and allocation of background pixels to the closest areas. The minimum size of the area is defined by the user. The source image of the sample of the aorta loaded from the file as well as user-defined and corrected (as a result of image splitting) boundaries of the area are displayed on the left side of the form as three superposed pixel layers. Using appropriate controls on the toolbar, Figure 6:Data processing manager. Karpenko et al.: Image processing31 Generally, the processing pipeline is a tree whose nodes are data processors. When the pipeline works, parent processors produce input data for child processors. The editing tools of the pipeline tree include embedding the selected processor in the pipeline as a parent node or a child node, removal of the selected node, or deletion of the subtree whose root is the selected node. Data filtering This FormView is an interface of one of the processors. Filtering is done by aortic identifiers and attributes as well as constraints on attribute values (see below). Setting constraints This FormView is a subform of the previous form and is used to set constraints on numeric values of attributes. Generally, constraints are specified as a set of upper and lower limits of the finite number of intervals of real numbers. General settings This FormView is meant for setting and changing application settings that are common to all users. Adjustable parameters are some default user names, databases, MySQL servers, and a port number. These data are stored in XML format in a special directory of the operating system. Conclusions This article proposes a method of automated research of aortic areas with atherosclerotic lesions in order to find a relationship between somatic mutations of the mitochondrial genome in the aortic wall cells and the extent of atherosclerotic lesions of these areas. The method includes the following basic stages: image segmentation of the aorta, labeling of connected components in an image, graphing region adjacency of the image, and adjacency graph partitioning. The authors proposed a structure; they designed and developed the AORTA system, which implements the proposed methodology. Testing and approbation of the AORTA system in the Russian Cardiology Research and Production Complex showed its workability and usability. The developed approach should lead to identification of those mtDNA mutations associated with atherosclerosis for their further use as novel genetic markers of individual predisposition to atherosclerosis. If some mutation is defined as atherogenic, its presence in mtDNA in other cells and tissues of the living subject (e.g., in blood monocytes, which migrate in the arterial wall and participate in the processes of atherogenesis) may define a higher likelihood of atherosclerosis development in the individual. The AORTA system has been demonstrated to have practical implications in the recent study published by Sazonova et al. [21]. The aim of that study was an analysis of heteroplasmy level of 11 mitochondrial mutations (652delG, A1555G, C3256T, T3336C, 652insG, C5178A, G12315A, G13513A, G14459A, G14846A, and G15059A) in normal and atherosclerotic fragments of morphologically mapped aortic intima. In brief, 265 normal and atherosclerotic intimal samples taken five human aortas were investigated. The samples were classified according to morphological characteristics into different types of atherosclerotic lesions: fibrous plaques, lipofibrous plaques, early atherosclerotic lesions (fatty infiltration and fatty streaks), and unaffected intima. At this stage, the AORTA system was employed to confirm the boundaries between different types of lesions. After classification, mtDNA was extracted from intimal samples; mtDNA fragments containing mutations were amplified by PCR, and the level of heteroplasmy was measured by pyrosequencing. G12315A and G14459A mutations have been found to be significantly associated with atherosclerotic lesions, especially early lesions and lipofibrous plaques. Mutation C5178A was significantly associated with fibrous plaques, whereas A1555G and G14846A mutations negatively correlated with early lesions and lipofibrous plaques [21]. This study has demonstrated the potential usefulness of AORTA system for basic and applied studies of pathogenesis of atherosclerosis. In the development of the work, the authors plan the following: ­ optimization of the database interface to improve its performance ­ use of a graphics tablet instead of a mouse to select boundaries of areas of the aorta ­ automatic images recognition as the main way to enter data about the geometry of areas of the aorta Author contributions: All the authors have accepted responsibility for the entire content of this submitted manuscript and approved submission. Research funding: This study was supported in part of morphological studies by the Russian Scientific Foundation (grant no. 14-14-01038). Employment or leadership: None declared. Honorarium: None declared. 32Karpenko et al.: Image processing Competing interests: The funding organization(s) played no role in the study design; in the collection, analysis, and interpretation of data; in the writing of the report; or in the decision to submit the report for publication. 10. Papamarkos NA. Technique for fuzzy document binarization. Proc ACM Symp Document Eng 2001:152­6. 11. Wu K, Otoo E, Suzuki K. Optimizing two-pass connected-component labeling algorithms. Pattern Anal Appl 2009;12: 206­20. 12. Haralick RM. Some neighborhood operations. In: Onoe M, Preston K Jr, Rosenfeld A, editors. Real time-parallel computing: image analysis. New York: Plenum Press, 1981:11­35. 13. Suzuki K, Horiba I, Sugie N. Linear-time connected-component labeling based on sequential local operations. Comput Vis Image Underst 2003:89:1­23. 14. Rosenfeld A, Pfaltz P. Sequential operations in digital picture processing. J Assoc Comput Mach 1966;12:471­94. 15. Chang F, Chen C-J, Lu C-J. A linear time component-labeling algorithm using contour tracing technique. Comput Vis Image Underst 2004;93:206­20. 16. Shapiro L. Connected component labeling and adjacency graph construction. In: Topological algorithms for digital image processing. Amsterdam: Elsevier, 1996:1­31. 17. He L, Chao Y, Suzuki K, Wu K. Fast connected-component labeling. Pattern Recog 2009;42:1977­87. 18. Sterzhanov MB. Methodology of selecting connected components in line binary images. Minsk: Belarusian State University of Informatics and Radioelectronics (BSUIR), 2006:18. 19. Ohlander R, Price K, Reddy DR. Picture segmentation using a recursive region splitting method. Comput Graphics Image Process 1978;8:313­33. 20. Karypis G, Kumar V. Multilevel k-way partitioning scheme for irregular graphs. J Parallel Distrib Comput 1998;8: 96­129. 21. Sazonova MA, Sinyov VV, Barinova VA, Ryzhkova AI, Zhelankin AV, et al. Mosaicism of mitochondrial genetic variation in atherosclerotic lesions of the human aorta. BioMed Res Int 2014: article ID 825468. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Bio-Algorithms and Med-Systems de Gruyter

AORTA software system for evaluating individual predisposition to atherosclerosis on the basis of genetic and phenotypic markers

Loading next page...
 
/lp/de-gruyter/aorta-software-system-for-evaluating-individual-predisposition-to-zZs3Qnuvm5

References

References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.

Publisher
de Gruyter
Copyright
Copyright © 2015 by the
ISSN
1895-9091
eISSN
1896-530X
DOI
10.1515/bams-2014-0023
Publisher site
See Article on Publisher Site

Abstract

This article deals with the AORTA software system providing support for research activities to find molecular basis for further assessment of individual predisposition to atherosclerosis. These studies are aimed at finding a relationship between somatic mutations of the mitochondrial genome in the aortic wall cells and the extent of atherosclerotic lesions of the aorta. A morphologist selects these areas on an aortic tissue sample and describes them, so that within each area, deviation of the quantitative indicator of atherosclerosis severity (phenotypic marker) from the area average should be sufficiently small. Next, the frequency and severity indicators of somatic mutations of the mitochondrial genome (genetic markers) are measured for each area and then entered into the AORTA system. Keywords: aortic wall; atherosclerosis; genetic and phenotypic markers; image processing; somatic mutations. DOI 10.1515/bams-2014-0023 Received December 21, 2014; accepted February 2, 2015 Introduction Currently, automation of medical and biomedical research is an important scientific and technical task. Identification of patients' predisposition to various diseases is of special importance in this research. This article deals with the development of AORTA system providing support for research carried out in the Russian Cardiology Research and Production Complex and the Institute of General Pathology and Pathophysiology of the Russian Academy *Corresponding author: Anatoly Karpenko, Computer-Aided Design, Bauman Moscow State Technical University, 2-ya Baumanskaya ul. 5, Moscow 105005, Russian Federation, E-mail: apkarpenko@mail.ru Valery Kharlamov: MSTU n.a. Bauman, Moscow, Russian Federation Igor Sobenin: Russian Cardiology Research and Production Complex, 3-ya Cherepkovskaya 15-a, 121552, Moscow, Russian Federation; and Institute of General Pathology and Patophysiology, Baltiyskaya ul., 5, 125315, Moscow, Russian Federation of Medical Sciences. These studies are aimed at finding a relationship between somatic mutations of the mitochondrial genome in the aortic wall cells and the extent of atherosclerotic lesions of the aorta [1]. Mitochondrial mutations were identified in various pathologies, such as coronary stenosis, some forms of type 2 diabetes mellitus and deafness, atherosclerosis, myocardial infarction, and cardiomyopathy. The mitochondrial genome is notable for its pronounced instability; therefore, frequent somatic mutations arise during the lifetime of an individual. Penetrance and expression of such mutations depend mainly on the genotype and the level of heteroplasmy. The latter represents a proportion of the number of the mitochondrial DeoxyriboNucleic Acid (mtDNA) copies bearing mutation (single nucleotide substitution, insertion, or deletion) at some position of the total number of mtDNA copies. Therefore, when studying the association of mitochondrial mutations with human disease, not only qualitative assessment (the presence/ absence of mutation) but also quantitative assessment of the mutant allele of the mitochondrial genome (the level of heteroplasmy) is necessary. To determine the critical level of heteroplasmy of mitochondrial mutations associated with formation and development of pathologies in humans, a method for direct quantitative assessment of the mitochondrial genome was developed [1]. As initial data for studying the role of mtDNA polymorphisms in the formation of atherosclerotic lesions of human arteries, the AORTA system uses quantitative severity of atherosclerotic lesions and mitochondrial mutations, as well as geometric shapes of the aorta areas under study. A morphologist selects these areas on an aortic tissue sample and describes them as unaffected regions, initial lesions, fatty streaks, lipofibrous, or fibrous atherosclerotic plaques, so that within each area, deviation of the quantitative indicator of atherosclerosis severity (phenotypic marker) from the area average should be sufficiently small. Next, total DNA is extracted from each morphologically identified aortic intimal sample, and the level of heteroplasmy for certain mtDNA mutations is measured in a quantitative manner. Therefore, the frequency and severity indicators of somatic mutations of the mitochondrial 26Karpenko et al.: Image processing genome (genetic markers) are measured for each area and then entered into the AORTA system. Analogues of the AORTA system are unknown to the authors. together with the results of their processing, and also, acting in the automatic mode, finds the boundaries of areas that have homogeneous lesions. The users should be able to associate the extensible set of properties with each of the areas: phenotypic markers, genetic markers, etc. Image processing must be carried out using an extensible set of processing modules. The sequence of using these modules must be user-defined. Problem statement The general view of the samples of human aorta derived from autopsy study is shown in Figure 1. In this figure, we can see selected areas within each of which the phenotypic marker (namely, the extent of atherosclerotic lesion) deviations from the area average are sufficiently small. The areas having a greater extent of atherosclerotic lesions appear saturated red. The areas are numbered consecutively, and regardless of the extent of atherosclerotic lesion, a unique ID number is provided for each sample. In its database, the AORTA system stores the images of aortic fragments that have atherosclerotic lesions, Methodology of solving the problem The proposed methodology of solving the problem consists of the following main stages: 1. image segmentation. 2. connected-component search and labeling. 3. graphing region adjacency of the labeled image. 4. adjacency graph partitioning. Image segmentation Image segmentation can be defined as a process of assigning labels to each pixel in the image so that sets of pixels with the same labels have common visual characteristics. Image segmentation results in a set of segments covering the image or a set of contours of these segments [2­4]. The following segmentation methods are best known: cluster analysis methods, including the iterative k-means algorithm [3]; histogram methods, which are non-iterative clustering methods [3]; methods of separation and fusion of areas [5]; area growth methods [6]; edge detection methods [5]; segmentation methods based on the morphological watershed algorithm, using an image as a three-dimensional surface defined by two spatial coordinates and a brightness level as surface elevation (relief) [5]; methods based on graph theory, for example, methods of normalized cut [7] and segmentation by weighted aggregation [8]; neural network methods [9, 10]. Our experiments with the methods of image segmentation showed that none of them could perform, with satisfactory accuracy, either automatic or automated selection of boundaries of the areas of aortic intima differing by the extent of atherosclerotic lesion under study. Therefore, it was decided that the user marks out the image manually and the AORTA system reads the marked image and automatically selects area boundaries. To solve the latter problem, we use global threshold binarization in one-dimensional attribute space by manually adjusting Figure 1:Photos of the aorta tissue samples (the areas that have homogeneous atherosclerotic lesions are selected). Karpenko et al.: Image processing27 the threshold. As an identifier, pixel brightness is used for monochrome images, and a scalar function of pixel color components is used for color images. Selection (marking) of connected components in an image B(r, c){0; 1} denotes the value of the pixel located at the intersection of the row r and the column c of the pixel array formed at the previous stage. We introduce a 4-connected N4(r, c) and an 8-connected N8(r, c) neighbors of this pixel. We assume that pixel (r, c) is the N4-neighbor to pixel (r, c), if (r, c)N4(r, c). Similarly, we define the N8-neighbor of pixel (r, c) to pixel (r, c). Pixel (r, c) is called the N4-connective to pixel (r, c) by the value v{0; 1} if there exists such a sequence of pixels (ri, ci), i=0, 1, 2, ..., n, (r0, c0)=(r, c), (rn, cn)=(r, c) that pixel (ri­1, ci­1)N4(ri, ci) and B(ri, ci)=v for all i[1:n]. The given sequence of pixels (ri, ci), i=0, 1, 2, ..., n, forms the connected N4-path from pixel (r, c) to pixel (r, c). Similarly, we define the N8-path from pixel (r, c) to pixel (r, c). The connected component of the pixel array B with the value v is formed by a set of C pixels having the value v, each pair of which is connected by the value v [11]. In the given symbols, labeling the connected components of the binary array B is as follows: select all the connected components in the given array and form a labeled image in which each pixel in the image is assigned a label (unique identifier) of the connected component to which the given pixel belongs. Allocation algorithms of connected components are divided into the following five classes [11]: multipass algorithms, two-pass algorithms, single-pass algorithms, algorithms using hierarchical structures of image representation, and parallel algorithms. Multipass algorithms use multiple scanning of images [12, 13]. Two-pass algorithms are classic allocation algorithms of connected components [14]. Single-pass algorithms form a class of the simplest connected components labeling algorithms. A single-pass algorithm scheme is as follows: 1. Find an unlabeled pixel with the value v=1 and assign it a new label. 2. Call the search procedure for all unprocessed neighbors of the given pixel with the same value v=1. 3. For each of the neighbors found, produce a recursive call of the search procedure and so on. The algorithm's search procedure can be constructed basing on the turn-and-stack data structures. In the first case, the algorithm is called breadth-first search algorithm, and in the second case, it is called depth-first search (DFS) algorithm. One of the most efficient singlepass algorithms is the contour tracing algorithm proposed by Chang et al. [15]. There is a highly effective image representation method called run-length encoding [16­18]. Due to high effectiveness and ease of implementation, the single-pass DFS algorithm having complexity O(n), where n is the number of pixels in the image, is used in the AORTA system. Graphing region adjacency of the labeled image Two regions of the image are called adjacent if some pixel in one of the regions is adjacent to some other pixel in another region. Information on related regions of the image is usually represented as region adjacency graph (RAG), where each node corresponds to a region of the image and the edges connect pairs of nodes corresponding to the adjacent regions. The AORTA system uses the algorithm for graphing region adjacency presented in the paper [3]. The algorithm processes the image by taking two lines at a time, showing horizontal and vertical adjacency, and if the N8-neighbor is considered, it also shows diagonal adjacency. When a new region adjacency is found, a new edge is added to RAG. Graph partitioning Suppose G=V, E is a given undirected weighted graph, where V is a vertex set and E is a set of edges. Graph partitioning is as follows: it is necessary to divide set V by k of the subsets Vi, i[1: k] so that any two subgraphs Vi, Vj, ij have no common vertices and that each of the vertices of the graph G belongs to one of the subgraphs. Graph partitioning is usually subordinate to the extremalization of some partitioning quality criterion. Therefore, we will talk about an optimal graph partitioning. This problem is known to be NP-hard. On this basis, in computation, we usually use the following heuristic methods of problem solving. The Kernighan-Lin algorithm is used to improve initial partitioning of the graph by sharing vertices among subgraphs so as to reduce the total number of cut edges [18]. Levelized nested dissection (LND) is used for finding a bisection of the graph that provides equality of the numbers of nodes in resulting subgraphs [19]. The spectral graph bisection algorithm uses two partitioning optimality criteria ­ minimal cut edges number and minimal difference between vertex numbers in different 28Karpenko et al.: Image processing subgraphs. The algorithm is based on the analysis of the eigenvector corresponding to the second largest eigenvalue of the original graph's Laplacian matrix [3]. Multilevel k-way partitioning [20] provides graph partitioning in three stages: coarsening of the original graph, partitioning the coarsened graph into k subgraphs, and recovery of the original graph. The latter algorithm is used for graph partitioning in the AORTA system. The database interface program includes (Figure 3) a data editor and a data processing subsystem. The editor allows the user to load data into the database, perform visualization of the data that are already in the database, modify the data, and submit the modification results to the database. The processing subsystem is a set of independent plug-ins for data processing as well as a shell that provides sharing of the modules. The processing subsystem also implements the user interface. We use MySQL as a database management system (DBMS). The main reason for this choice is to support a large number of MySQL table types, for example, MyISAM tables for full text search and InnoDB tables for transactions at the level of individual records. The application was developed on the.NET Framework 3.5 platform using the C# programming language. Microsoft Visual Studio 2008 was used as an integrated development environment; it has advanced graphical user interface programming aids and provides ample opportunities for project management and debugging. MySQL Connector/NET library was used to interact with the MySQL DBMS. This library implements all the necessary interfaces of ADO.NET, which is part of the.NET Framework providing the programmer with uniform access to different data sources. The Windows Forms library was used to create the graphical user interface. The AORTA bundled software Structure of the AORTA bundled software The structure of the AORTA bundled software is shown in Figure 2. The bundled software consists of a relational database and its software interface, that is, the front end of the system. < > < > : Workstation < > : Database server < > : .NET Framework < > : MySQL Server < > : AORTA.exe : Database schema FormView controls TCP/IP 1 Figure 2:Structure of the AORTA system. FormView controls include "login", the main "AORTA" view, "data editor", "selecting image boundaries", "color < > Program system < > Data editor < > Set of plug-ins IDataStreamProcessor < > Data processing subsystem < > < > MySQL Connector/NET < > .NET Framework < > : AORTA.exe Figure 3:Diagram of components. Karpenko et al.: Image processing29 gradient assignment", "data processing manager", "data filtering", "setting constraints", and "general settings". Login This FormView control is used to set the username and password for further authentication; it also contains fields for setting network address and a port to connect to a MySQL Server Instance (these fields are available only when the application is run under administrator account). sample regions on which relative character values of the regions are indicated by means of color gamut. To configure the imaging parameters, a tool for setting a linear gradient is included in the graphics editor. The left side of the data editor FormView control contains tables to display text and numeric data. The right side displays a graphical image map of aortic regions selected for studying a sample of the aorta. At the bottom, there is a stack for displaying auxiliary data. Selecting image boundaries The FormView control (Figure 5) allows the user to select an area of the aorta. The user's work with this form begins with loading the image of the aorta sample from the file. Next, the user can manually select region boundaries (for example, using a mouse) or use built-in image segmentation. Manual selection of boundaries includes the brush tool and the eraser tool, the parameters of which can be modified using interface elements located in the toolbar on the right side of the form. As mentioned above, the built-in segmentation tools support threshold binarization of the loaded image with manual setting of intensity threshold. To help the user select the correct threshold, a pixel intensity histogram of the image is displayed in the toolbar. The result of setting image boundaries is a binary image of the aortic area under study. Pressing the refresh button implements image splitting ­ that is, it implements automatic search and labeling of the connected The main application window If authentication is successful, the user menu is available in the main application window. Accessibility (activation capability) to menu items is defined by the active user privileges. Data editor This FormView control (Figure 4) allows the user to input data into the database, editing, and visualization. To set geometric shapes of the aortic areas, a graphics editor allowing to set positions of region boundaries manually or use image segmentation tools is included in the data editor. In the latter case, threshold image binarization (item 1) is implemented and intensity threshold is set manually. Data visualization includes a map of aorta Figure 4:Data editor. 30Karpenko et al.: Image processing one can enable and disable display of each of the layers as well as image zoom. Color gradient assignment The color gradient assignment FormView allows the user to select one of the predefined color gradients or specify a custom gradient. Piecewise linear color gradient is presented in the form of the vector-function G(n)=(gR(n), gG(n), gB(n)), whose values are RGB color components. The argument of this function is a normalized color value nR, n[0; 1]. The value of the gradient is determined basing on piecewise linear interpolation of color components in userdefined color stops for which corresponding values n (in percentage terms) are defined. The user can change the number of color stops, their location, and color. Figure 5:Selecting image boundaries. Data processing manager This FormView (Figure 6) provides control tools for data processing sequence (pipeline). The left side of the form displays a list of available processing modules (processors), their characteristics, and description. Processors can have their own user interfaces; so, in general, the pipeline presupposes interaction with the user. The right side of the form allows the user to create new pipelines, select, and edit previously formed pipelines. This part of the form allows to run the selected pipeline in a singlestep or nonstop mode. components of the image, removing insufficiently large areas, and allocation of background pixels to the closest areas. The minimum size of the area is defined by the user. The source image of the sample of the aorta loaded from the file as well as user-defined and corrected (as a result of image splitting) boundaries of the area are displayed on the left side of the form as three superposed pixel layers. Using appropriate controls on the toolbar, Figure 6:Data processing manager. Karpenko et al.: Image processing31 Generally, the processing pipeline is a tree whose nodes are data processors. When the pipeline works, parent processors produce input data for child processors. The editing tools of the pipeline tree include embedding the selected processor in the pipeline as a parent node or a child node, removal of the selected node, or deletion of the subtree whose root is the selected node. Data filtering This FormView is an interface of one of the processors. Filtering is done by aortic identifiers and attributes as well as constraints on attribute values (see below). Setting constraints This FormView is a subform of the previous form and is used to set constraints on numeric values of attributes. Generally, constraints are specified as a set of upper and lower limits of the finite number of intervals of real numbers. General settings This FormView is meant for setting and changing application settings that are common to all users. Adjustable parameters are some default user names, databases, MySQL servers, and a port number. These data are stored in XML format in a special directory of the operating system. Conclusions This article proposes a method of automated research of aortic areas with atherosclerotic lesions in order to find a relationship between somatic mutations of the mitochondrial genome in the aortic wall cells and the extent of atherosclerotic lesions of these areas. The method includes the following basic stages: image segmentation of the aorta, labeling of connected components in an image, graphing region adjacency of the image, and adjacency graph partitioning. The authors proposed a structure; they designed and developed the AORTA system, which implements the proposed methodology. Testing and approbation of the AORTA system in the Russian Cardiology Research and Production Complex showed its workability and usability. The developed approach should lead to identification of those mtDNA mutations associated with atherosclerosis for their further use as novel genetic markers of individual predisposition to atherosclerosis. If some mutation is defined as atherogenic, its presence in mtDNA in other cells and tissues of the living subject (e.g., in blood monocytes, which migrate in the arterial wall and participate in the processes of atherogenesis) may define a higher likelihood of atherosclerosis development in the individual. The AORTA system has been demonstrated to have practical implications in the recent study published by Sazonova et al. [21]. The aim of that study was an analysis of heteroplasmy level of 11 mitochondrial mutations (652delG, A1555G, C3256T, T3336C, 652insG, C5178A, G12315A, G13513A, G14459A, G14846A, and G15059A) in normal and atherosclerotic fragments of morphologically mapped aortic intima. In brief, 265 normal and atherosclerotic intimal samples taken five human aortas were investigated. The samples were classified according to morphological characteristics into different types of atherosclerotic lesions: fibrous plaques, lipofibrous plaques, early atherosclerotic lesions (fatty infiltration and fatty streaks), and unaffected intima. At this stage, the AORTA system was employed to confirm the boundaries between different types of lesions. After classification, mtDNA was extracted from intimal samples; mtDNA fragments containing mutations were amplified by PCR, and the level of heteroplasmy was measured by pyrosequencing. G12315A and G14459A mutations have been found to be significantly associated with atherosclerotic lesions, especially early lesions and lipofibrous plaques. Mutation C5178A was significantly associated with fibrous plaques, whereas A1555G and G14846A mutations negatively correlated with early lesions and lipofibrous plaques [21]. This study has demonstrated the potential usefulness of AORTA system for basic and applied studies of pathogenesis of atherosclerosis. In the development of the work, the authors plan the following: ­ optimization of the database interface to improve its performance ­ use of a graphics tablet instead of a mouse to select boundaries of areas of the aorta ­ automatic images recognition as the main way to enter data about the geometry of areas of the aorta Author contributions: All the authors have accepted responsibility for the entire content of this submitted manuscript and approved submission. Research funding: This study was supported in part of morphological studies by the Russian Scientific Foundation (grant no. 14-14-01038). Employment or leadership: None declared. Honorarium: None declared. 32Karpenko et al.: Image processing Competing interests: The funding organization(s) played no role in the study design; in the collection, analysis, and interpretation of data; in the writing of the report; or in the decision to submit the report for publication. 10. Papamarkos NA. Technique for fuzzy document binarization. Proc ACM Symp Document Eng 2001:152­6. 11. Wu K, Otoo E, Suzuki K. Optimizing two-pass connected-component labeling algorithms. Pattern Anal Appl 2009;12: 206­20. 12. Haralick RM. Some neighborhood operations. In: Onoe M, Preston K Jr, Rosenfeld A, editors. Real time-parallel computing: image analysis. New York: Plenum Press, 1981:11­35. 13. Suzuki K, Horiba I, Sugie N. Linear-time connected-component labeling based on sequential local operations. Comput Vis Image Underst 2003:89:1­23. 14. Rosenfeld A, Pfaltz P. Sequential operations in digital picture processing. J Assoc Comput Mach 1966;12:471­94. 15. Chang F, Chen C-J, Lu C-J. A linear time component-labeling algorithm using contour tracing technique. Comput Vis Image Underst 2004;93:206­20. 16. Shapiro L. Connected component labeling and adjacency graph construction. In: Topological algorithms for digital image processing. Amsterdam: Elsevier, 1996:1­31. 17. He L, Chao Y, Suzuki K, Wu K. Fast connected-component labeling. Pattern Recog 2009;42:1977­87. 18. Sterzhanov MB. Methodology of selecting connected components in line binary images. Minsk: Belarusian State University of Informatics and Radioelectronics (BSUIR), 2006:18. 19. Ohlander R, Price K, Reddy DR. Picture segmentation using a recursive region splitting method. Comput Graphics Image Process 1978;8:313­33. 20. Karypis G, Kumar V. Multilevel k-way partitioning scheme for irregular graphs. J Parallel Distrib Comput 1998;8: 96­129. 21. Sazonova MA, Sinyov VV, Barinova VA, Ryzhkova AI, Zhelankin AV, et al. Mosaicism of mitochondrial genetic variation in atherosclerotic lesions of the human aorta. BioMed Res Int 2014: article ID 825468.

Journal

Bio-Algorithms and Med-Systemsde Gruyter

Published: Mar 31, 2015

References