Automatic localization of anatomical regions in medical ultrasound images of rheumatoid arthritis using deep learning

Abstract

The pace of population aging is growing faster worldwide. The quality of life of the aging population is mostly affected by rheumatic diseases. With the increasing rate of rheumatoid arthritis in the aging population, technological advances in the field of automatic image processing and analysis have paved way for automatic detection and diagnosis of arthritis based on how the grade of the synovial region is designed. The proposed method is based on spatial analysis using intensity-based approach to segment the skin border, thresholding and connectivity algorithm for bone region segmentation, hit-or-miss transform for bone line segmentation and distance measure with image profile to detect the joint region. After this process of localization, the synovial region is determined using the active contour technique. In arthritis condition, synovitis also occurs which is categorized into four different grades based on the fluid expansion in the synovial region. The different grades are defined and analyzed through deep learning. Convolutional neural network in a deep learning algorithm is used to diagnose the particular grade of synovitis to describe the nature of arthritis. With these results, a module to detect the nature of arthritis automatically is defined.