Atherosclerotic plaque classification in carotid ultrasound images using machine learning and explainable deep learning

• DOI: 10.1016/j.imed.2023.05.003
• Publication Date: 2023-07-01T06:13:16Z
• Authors: Soni Singh; Pankaj K. Jain; Neeraj Sharma; Mausumi Pohit; Sudipta Roy
• Keywords: Interpretability

The incidence of cardiovascular diseases (CVD) is rising rapidly worldwide. Some forms CVD, such as stroke and heart attack, are more common among patients with certain conditions. Atherosclerosis development a major factor underlying events, attack stroke, its early detection may prevent events. Ultrasound imaging carotid arteries useful method for diagnosis atherosclerotic plaques; however, an automated to classify plaques evaluation early-stage CVD needed. Here, we propose classification high-risk plaque ultrasound images. Five deep learning (DL) models (VGG16, ResNet-50, GoogLeNet, XceptionNet, SqueezeNet) were used the results compared those machine (ML)-based technique, involving extraction 23 texture features from images using Support Vector Machine classifier. To enhance model interpretability, output gradient-weighted convolutional activation maps (gradCAMs) generated overlayed on original A series indices, including accuracy, sensitivity, specificity, F1-score, Cohen-kappa index, area under curve values, calculated evaluate performance. GradCAM allowed visualization most significant image regions. Of tested, GoogLeNet yielded highest accuracy (98.20%). ML also suitable applications requiring low computational resource, their (96.85%) similar that DL models. Further, can be completely than