Ultrasound Nerve Segmentation with Deep Learning for Leprosy

Purpose: This study aims to provide an AI tool for detecting nerves in ultrasound images to help diagnose Hansen's disease (Leprosy) in rural areas. The significant difference in the cross-sectional area (CSA) of superficial nerves in symmetrical extremities is a landmark in the early stages of the disease. Despite its potential, ultrasound nerve evaluation is limited due to the difficulty in accurately identifying nerves in ultrasound images. Methodology: We propose the first Leprosy video nerve segmentation pipeline based on YOLOv8 and X-Mem architectures to automate frame detection, segmentation, and label propagation. We ensure alignment with clinical practices and evaluate the inference in real time of the method and its energy efficiency, confirming the approach's feasibility in resource-limited settings. Results: We establish a baseline for nerve segmentation of ultrasound Leprosy videos, presenting the first results to identify relevant frames, segment, and propagate labels. To support further research, we have open source a new leprosy test dataset and created a demo web page to try our method on real patient data ¹¹http://github.com/vgonzalezd/Leprosy. This initiative aims to promote research on AI techniques to improve healthcare in rural communities, where healthcare professionals are scarce and assistance is essential.