Performance Analysis of Machine Learning and Deep Learning Architectures on Early Stroke Detection Using Carotid...

Area of Science:

• Cardiovascular Imaging
• Medical Artificial Intelligence
• Biomedical Engineering

Background:

• Cardiovascular diseases (CVDs) pose a significant global health burden.
• Early detection of CVDs is crucial for effective management and prevention.
• Atherosclerotic plaque in the carotid artery serves as an important indicator for CVD risk.

Purpose of the Study:

• To evaluate the efficacy of machine learning (ML) and deep learning (DL) algorithms in classifying carotid artery ultrasound images.
• To differentiate between symptomatic and asymptomatic individuals based on carotid artery plaque characteristics.
• To assess the performance of various ML and DL models in identifying carotid artery stenosis.

Main Methods:

• A dataset of 361 carotid artery ultrasound images was utilized, comprising 202 normal images and 159 images with carotid plaque.
• Machine learning algorithms including CART decision tree, random forest, and logistic regression were applied.
• Deep learning models such as Convolutional Neural Network (CNN), Mobilenet, and Capsulenet were implemented for image classification.

Main Results:

• The random forest algorithm achieved a classification accuracy of 91.41%.
• The Capsulenet transfer learning approach demonstrated a high accuracy of 96.7% in classifying the images.
• Both ML and DL methods showed promising results in identifying carotid artery plaque and assessing patient status.

Conclusions:

• Machine learning and deep learning techniques are effective tools for analyzing carotid artery ultrasound images.
• Capsulenet, in particular, shows high potential for accurate classification of carotid artery ultrasound data.
• These AI-driven approaches can aid in the early identification of cardiovascular disease risk by classifying patients as symptomatic or asymptomatic.