A Hybrid Approach for Optimal Multi-Class Classification of Neglected Tropical Skin Diseases using Multi-Channel HOG Features

Abstract

Neglected tropical skin diseases (NTDs) pose significant health challenges, especially in resource-limited settings. Early diagnosis is crucial for effective treatment and preventing complications. This study proposes a novel multi-class classification approach using multi-channel HOG features and a hybrid metaheuristic algorithm to improve the accuracy of NTD diagnosis. The method extracts optimal HOG features from images of Buruli Ulcer, Leprosy, and Cutaneous Leishmaniasis through different cell sizes, generating multiple training datasets. A hybrid Whale Optimization Algorithm and Shark Smell Optimization Algorithm (WOA-SSO) optimizes the Error Correcting Output Code (ECOC) framework for SVM, achieving superior multi-class classification performance. Notably, the multi-channel dataset, derived from averaging HOG features of different cell sizes, yields the highest accuracy of 89%. This study demonstrates the potential of the proposed method for developing mobile applications that facilitate early and accurate diagnosis of NTDs through image analysis, potentially improving patient outcomes and disease control. The hybrid metaheuristic algorithm plays a crucial role in optimizing the ECOC framework, enhancing the accuracy and efficiency of the multi-class classification process. This approach holds significant promise for revolutionizing NTD diagnosis and management, particularly in underserved communities.