Classification of calcified regions in atherosclerotic lesions of the carotid artery in computed tomography angiography images

• Published in: Neural computing & applications 2020-04, Vol.32 (7), p.2553-2573
• Main Authors: Jodas, Danilo Samuel, Pereira, Aledir Silveira, Tavares, João Manuel R. S.
• Format: Article
• Language: English
• Subjects: Angiography; Artificial Intelligence; Atherosclerosis; Calcification; Carotid arteries; Classification; Classifiers; Computational Biology/Bioinformatics; Computational Science and Engineering; Computed tomography; Computer Science; Correlation coefficients; Data Mining and Knowledge Discovery; Diagnostic software; Diagnostic systems; Feature extraction; Image classification; Image Processing and Computer Vision; Medical imaging; Morphology; Original Article; Pixels; Probability and Statistics in Computer Science; Veins & arteries
• ISSN: 0941-0643; 1433-3058
• DOI: 10.1007/s00521-019-04183-z

The identification of atherosclerotic plaque components, extraction and analysis of their morphology represent an important role towards the prediction of cardiovascular events. In this article, the classification of regions representing calcified components in computed tomography angiography (CTA) images of the carotid artery is tackled. The proposed classification model has two main steps: the classification per pixel and the classification per region. Features extracted from each pixel inside the carotid artery are submitted to four classifiers in order to determine the correct class, i.e. calcification or non-calcification. Then, geometrical and intensity features extracted from each candidate region resulting from the pixel classification step are submitted to the classification per region in order to determine the correct regions of calcified components. In order to evaluate the classification accuracy, the results of the proposed classification model were compared against ground truths of calcifications obtained from micro-computed tomography images of excised atherosclerotic plaques that were registered with in vivo CTA images. The average values of the Spearman correlation coefficient obtained by the linear discriminant classifier were higher than 0.80 for the relative volume of the calcified components. Moreover, the average values of the absolute error between the relative volumes of the classified calcium regions and the ones calculated from the corresponding ground truths were lower than 3%. The new classification model seems to be adequate as an auxiliary diagnostic tool for identifying calcifications and allowing their morphology assessment.