Ultrasound lmaging-vulnerable plaque diagnostics: Automatic carotid plaque segmentation based on deep learning

Abstract

PurposeThe purpose of this article is to develop a novel algorithm based on a kind of deep convolutional neural network for the automatic carotid plaque segmentation and stroke risk assessment from the carotid artery plaque using Doppler ultrasound system. This can improve the segmentation accuracy of carotid artery plaque from ultrasound scans which enhance the accuracy of stroke risk assessment caused by carotid plaque, and help researchers and experts to make more accurate quantitative estimates. MethodThis study used the Doppler system, combined with segmentation technique, to identify the nature and structure of plaques, which could be used for risk assessment of people at high risk of clinical suspicion of stroke. The carotid Doppler scan images used in this experiment were mainly from patients in the radiological department of a hospital. A total of 568 carotid ultrasound scanning of carotid Plaque were included. This experiment uses ResNet50, Inception_v3, and RPN for feature extraction of medical images, using RELU as the activation function. The data set after data enhancement is used for training and prediction, and the experimental results are compared. ResultBased on the physiological analysis of method we used and the risk assessment, we achieved results consistent with the result we predicted by doing a series of experiments and comparing different groups. We implemented a deep learning framework to segment Doppler ultrasound images, and achieved the highest accuracy of 92.94% by using fine-tuned and adjusted hyperparameters. ConclusionResults suggested that the ResNet50 and Inception_v3 may be used to classify the types of plaque and the method we used was more accurate than the traditional methods, which indicated that we just found an appropriate way to analyze plaque vulnerability and find the solution in time. It can provide a reliable and flexible method in recent clinical drug interventions.