Optimized real-time diagnosis of neglected tropical diseases by automatic recognition of skin lesions

Background Skin lesions associated with certain endemic neglected tropical diseases found in landlocked areas have irreversible effects on populations, contributing to population impoverishment. According to the World Health Organization (WHO), there were 127558 new leprosy cases detected in 2020, with about 8 629 children under the age of 15 affected with approximately 7 198 suffering from grade 2 disabilities (G2D). In the same year, about 1 million new leishmaniasis cases were reported globally, whereas the number of suspected Buruli ulcer cases reported globally in one year was around 5000, with the majority of patients under the age of 15. Current diagnostic methods are unfortunately expensive, not suitable for endemic areas, and sometimes not very effective. According to the SDG Target 3.3, WHO plans to end the neglect tropical skin diseases (NTDs) by 2030. According to them, this required an early diagnosis. Our work belongs to this objective. Methods This paper presents an optimized diagnostic approach for neglected tropical skin diseases (NTDs) by automatically identifying skin lesions in their early phase. We chose three main skin diseases with similar early lesion development. First of all, we have built a dataset of images of these lesions. After accurately detecting the location of the skin lesion, we delineate the edges of the lesion. In addition, we apply a series of filters to detect and extract relevant features from these lesions. Finally, the detected features are fed into a support vector machine (SVM) optimized by a black hole algorithm (BHO) to be classified in real time (less processing time). Results Our model achieves a global classification accuracy of 96%, 94% specificity, an F-SCORE of 89%, a recall of 90%, and a sensitivity of 92% of the images under different conditions and a processing time that is lower than other classical algorithms. These results are superior to current diagnostic methods such as direct smear examination with a low sensitivity (<60%), in vitro culture with a low sensitivity (20–60%), and PCR for IS2404 with high sensitivity and specificity (>90%) (but which requires sophisticated laboratories and competent personnel). These results can improve the sanitary situation of the population and enable the development of efficient Computer-Aided Diagnosis (CAD). Conclusion The results obtained in this research produce the best solution compared to other algorithms and can be used to make a real time diagnosis of NTDs. Trial registration “Not applicable”