Automatic Detection and Quantification of Carotid Atherosclerotic Plaque Parameters from B-mode Ultrasound Images Using Deep Learning

Carotid atherosclerosis is a major cause of cerebrovascular events, including ischemic strokes. B-mode ultrasound (US) is a safe, widely available, and cost-effective imaging modality used to detect and assess atherosclerotic plaques. However, it has limitations, namely, high inter-operator and skill-dependent variability in interpretation. We developed a time-efficient deep convolutional neural network (CNN)-based workflow to automatically 1) detect and segment plaques from US (longitudinal/transverse), and 2) quantify plaque parameters: thickness (T), total plaque area (TPA), and total plaque volume (TPV). Patients (n=141) with severe carotid atherosclerotic plaques underwent US examination prior to carotid endarterectomy. A radiologist annotated US images (467), longitudinal (326) and transverse (141). The Intersection Over Union between manual and algorithm-based segmentation was 0.80±0.18 (mean±standard deviation) on 46 unseen US images. The mean absolute error of (∆T), (∆TPA), and (∆ V) were 0.22 mm, 0.73 mm
²
, and 0.89 mm
³
. The total time (detection/quantifications) was <1 second per image.