Leprosy Early Detection Through Binary Segmentation Using ResU-Net

Abstract

Leprosy is a chronic infectious disease caused by Mycobacterium leprae that can lead to physical deformity if left untreated. Indonesia currently ranks third in the world for leprosy prevalence, with the highest concentration of cases found in the provinces of West Papua, North Maluku, and Papua. These provinces, located in the eastern region of Indonesia, face numerous challenges in terms of healthcare accessibility for early leprosy detection due to various factors and novel, more accessible method to detect leprosy is urgently needed. In this study, we introduce an innovative approach to early leprosy detection by leveraging the ResU-Net model. The ResU-Net, a hybrid architecture, combines the robust U-Net framework, renowned for its efficacy in medical image segmentation, with the powerful ResNet-50 and ResNet-101 backbones. The incorporation of ResNet-50 and ResNet-101 enhances the model's capability to extract intricate features from the target image, allowing for a more comprehensive analysis and ultimately, a more accurate and early detection of leprosy. To train and validate our model, we employ the CO2Wounds Leprosy dataset, a comprehensive collection of medical images showcasing images of leprosy taken using smartphone. The research results demonstrate the promising potential of ResU-Net in accurately identifying leprosy-affected areas within these images with highest IoU scores of around 80% with the ResNet101 backbone and around 79% with the ResNet50 backbone. This method holds great potential for improving the management of leprosy in regions with high prevalence by enabling accessible and timely interventions.