Abstract
This study is dedicated to designing and advancing machine learning (ML) algorithms for classifying normal and abnormal muscular tissues, thereby aiding neurologists in diagnosing inclusion body myositis (IBM). Our work mainly aims to leverage machine learning and recent state-of-the-art (SOTA) algorithms to recognize and diagnose myositis from muscle ultrasound images in the preliminary stage and support the traditional diagnostic methodology. Initially, we used an open-source ultrasound image dataset to construct and refine initial models using VGG-16. We employed the Grad-CAM method to annotate muscle ultrasound images and delineate regions of interest (ROI). Subsequent experiments enhanced the VGG16 architecture through extensive layer modifications and parameter adjustments. Our research offers valuable perspectives on utilizing ML to assist neurologists in the early diagnosis of IBM.