Key points are not available for this paper at this time.
Skin neglected tropical diseases (NTDs) remain a major public health challenge in low- and middle-income countries, where frontline health workers (FHWs) often lack dermatological training. In response, the World Health Organization (WHO) created the Skin NTDs App-developed by UniversalDoctor-to support FHWs in resource-limited settings. Initially created as a digital adaptation of a WHO's training guide, the App evolved by incorporating another clinical decision support tool (CDST) from until No Leprosy Remains and an artificial intelligence (AI)-powered visual classifier (VC). Our purpose is to describe the WHO Skin NTDs App and evaluate its AI-powered VC. The VC was trained to identify 12 skin NTDs out of 13 through a convolutional neural network (DenseNet-121). Performance was assessed through sensitivity, specificity, accuracy, precision, F1-score, and per-class sensitivity from top-1 through top-5. The VC demonstrated high performance in internal evaluations, achieving 99.8% top-5 sensitivity across all diseases and top-1 accuracy above 75% for most diseases. Some underrepresented conditions (e.g., chromoblastomycosis, sporotrichosis) showed lower precision. In conclusion, the AI-powered WHO Skin NTDs App is a promising digital tool for capacity-building of FHW in underserved areas. Continued development, external validation, and integration into clinical workflows will be critical to assess its performance globally.
Pons et al. (Fri,) studied this question.