Early Detection and AI-Based Management of Skin-Related Neglected Tropical Diseases in Sub-Saharan Africa by Frontline Health Workers

Overview

Skin-related Neglected Tropical Diseases (Skin NTDs) affect about 1.8 billion people worldwide, particularly in poor and rural communities where healthcare access is limited. Many people rely on frontline health workers (FHWs) for treatment, but these workers often lack specialized training in skin diseases, making diagnosis difficult. To address this challenge, the SkincAIr project is testing whether a mobile app powered by artificial intelligence (AI) can help FHWs improve their ability to detect Skin NTDs. The study will be conducted in two arms. In the first clinical image data collection arm (36 months), dermatologists in 5 countries (Kenya, Ethiopia, Senegal, Democratic Republic of Congo and Nigeria) will collect images of skin NTD and other skin conditions that will be used for development and training of the AI model within the SkincAIr app before it is tested among FHWs. The second validation study arm will take place in 3 countries (Kenya, Ethiopia and Senegal), and will involve 50 FHWs and around 750 patients in each country over 24 months. During the first 12 months (Phase A), FHWs will diagnose patients using standard methods without the app, establishing baseline performance on key indicators including diagnostic accuracy, time to diagnosis, referral patterns, and cost implications of improved primary-level diagnosis. For the following 6 months (Phase B), FHWs will use the SkincAIr app with AI functionality activated to support diagnosis and enable real-time geolocated disease mapping and hotspot identification. In the final 6 months (Phase C), the app is withdrawn to assess whether FHWs retain their improved diagnostic skills. We will summarize the results using simple numbers and charts to show how often things happen and what the average results look like. Researchers will evaluate how well the app improves diagnosis by FHWs and whether FHWs retain their improved skills even after AI support is removed, by comparing their results with those of a skin specialist (dermatologist). Interviews and group discussions will be recorded, written down, organized into key ideas, and carefully reviewed using a computer program to understand the main themes. Study findings will be shared with National Ministries of Health, presented at local and international conferences, and reported to relevant institutional and regulatory authorities. If successful, this AI tool could boost early detection of skin diseases, enhance disease tracking, and improve healthcare in underserved areas.

ABSTRACT: Skin-related Neglected Tropical Diseases (Skin NTDs) pose a significant public health challenge, affecting 1.8 billion people globally. Skin NTDs significantly affect marginalized communities due to several factors, such as lack of trained healthcare staff and diagnostic tools. Currently, due to a scarcity of dermatologists, the majority of the rural population with skin diseases is served by frontline health workers (FHWs) with limited dermatological knowledge. The low prevalence of skin NTDs further confounds their diagnosis and recognition by FHWs. Novel innovative approaches are therefore needed to build capacity and improve diagnosis for skin NTDs. Mobile health (mHealth) interventions, particularly those incorporating artificial intelligence (AI), offer a promising solution to enhance the diagnostic capabilities of FHWs. The SkincAIr project aims to evaluate whether introducing a mobile app with AI functionality can improve the diagnostic accuracy (sensitivity and specificity) of FHWs in detecting skin NTDs, and to determine their retention of improved diagnostic skills after the AI assistance is removed, indicating potential capacity building and sustained improvement in healthcare delivery. In the clinical image data collection arm (36 months), dermatologists in 5 countries (Kenya, Ethiopia, Senegal, Democratic Republic of Congo and Nigeria) will collect images of skin NTD and other skin conditions (Image data collection phase) that will be used for training and development of the SkincAIr app, before it is evaluated among FHWs during the 24-month validation study. The validation study arm for the app will involve a within-subjects longitudinal design which will enroll 50 FHWs and will recruit about 750 patients with skin complaints from areas with high burden of skin NTDs in each of 3 countries (Kenya, Ethiopia, Senegal) over a 24-month validation study period. Data will be analyzed using R and Python, with descriptive statistics (frequency, central tendency, and dispersion) summarized in tables and charts. Diagnostic accuracy of FHWs before and after app introduction will be evaluated using sensitivity, specificity, predictive values, and percent agreement with a dermatologist. Qualitative data from interviews and FGDs will be audio-recorded, transcribed, coded, and thematically analyzed using Atlas.ti Version 7. Study findings will be shared with National Ministries of Health, presented at local and international conferences, and reported to IRBs and regulatory authorities. It is envisaged that the app will improve the diagnostic accuracy of FHWs in early detection of skin NTDs and will facilitate real-time epidemiological surveillance, contributing to improved disease mapping and hotspot identification. INTRODUCTION/BACKGROUND: Skin-related Neglected Tropical Diseases (MDPI, 2019) (skin NTDs) such as leprosy, Buruli ulcer, yaws (endemic treponematosis), cutaneous leishmaniasis, chromoblastomycosis, mycetoma, scabies, tungiasis, Post Kala-azar Dermal Leishmaniasis (PKDL), lymphatic filariasis, onchocerciasis, podoconiosis and sporotrichosis pose a significant public health challenge, affecting 1.8 billion people globally at any given moment (WHO, 2023). Particularly in sub-Saharan Africa (SSA), these diseases are highly prevalent and are linked to substantial health inequities, predominantly impacting marginalized communities (Kariuki et al. 2023). Kenya for instance has a significant burden of Skin NTDs, including Lymphatic filariasis at the Coastal region (Njenga et al. 2017; Ofire et al. 2025), Mycetoma in Turkana (Colom et al. 2023), Leishmaniasis in Rift Valley and Eastern parts (Baringo, Naivasha/Gilgil, Laikipia, Samburu, Nakuru, Meru, West Pokot, Elgeyo Marakwet, Isiolo, Nyandarua, and Marsabit) (Ngere et al. 2020; van Dijk et al. 2024), Tungiasis (Elson et al. 2019; Nyangacha et al. 2019) and Scabies (Schmeller and Dzikus, 2001; Mbogori 2014; Macharia et al. 2024) that have a wide distribution across the country and some pockets of cases of Leprosy in Kwale, Kilifi, Kisumu, Siaya, Homabay and Busia counties (Kenya NTLLD Program Annual Report, 2014; Wangara et al. 2019). The prevalence of these diseases is exacerbated by factors such as poverty and a lack of adequate healthcare resources, notably insufficiently trained staff for effective management of skin NTDs (Ochola et al. 2021). The stigma surrounding skin NTDs, entrenched in societal and economic contexts, leads to isolation and discrimination, discouraging diagnosis or treatment, which in turn exacerbates disease spread and complicates control and elimination efforts, creating a self-perpetuating cycle of challenges. A high proportion of NTDs have major skin manifestations. Therefore, examination of the skin serves as an opportunity to identify multiple NTDs in a single intervention. The integrative approach, recommended by the World Health Organization (WHO) (https://www.who.int/activities/promoting-the-integrated-approach-to-skin-related-neglected-tropical-diseases), results in enhanced case detection and increased efficiency through sharing of resources and expanded programme coverage. However, there is a major barrier to the integration of skin NTD interventions: the lack of dermatologists (Schmid-Grendelmeier et al. 2019) (one dermatologist for 1-2 million inhabitants) and adequately trained healthcare staff. Currently, the majority of the rural population with skin diseases is served by frontline health workers (FHWs) with limited dermatological knowledge (Mieras et al. 2018). This challenge is further confounded by the low prevalence of skin-related NTDs, which makes them difficult to be recognized by FHWs (Mieras et al. 2018; Hotez et al. 2009). As skin NTDs rely mostly on clinical diagnosis, lack of adequate training of FHWs jeopardizes disease control programs and the attainment of the overall 2021-2030 WHO NTD roadmap target to reduce morbidity, disability and the psychosocial impact of skin NTDs by 2030. Novel innovative approaches are needed to build capacity and improve diagnosis for skin NTDs to realize the 2030 goals. In this regard, Artificial Intelligence (AI) now provides an unprecedented opportunity to use advances in medical imaging applied to the skin to tackle current barriers in the diagnosis and management of skin NTDs in SSA. Existing AI models in dermatology often focus on diseases prevalent in developed countries and rely on homogeneous datasets, leading to models that do not generalize well to diverse populations. They typically use internal validation methods, which are insufficient for real-world deployment where models encounter varied data sources (Daneshjou et al. 2021). Whereas AI-powered algorithms have demonstrated diagnostic accuracy similar to expert clinicians in high-resource settings (Salinas et al. 2024), very few studies have explored the use of AI-powered apps for diagnosing skin NTDs in low-resource settings in SSA. This is due to the novelty of the technology which contributes to the scarcity of such research. The WHO NTD-led Global Initiative discusses progress, challenges, gaps and solutions in developing and implementing artificial augmented intelligence-based apps as a capacity building and monitoring tool for skin NTDs and selected common skin conditions in resource-limited settings. Recently, the WHO incorporated two online AI algorithms that intend to classify 12 skin NTDS and 24 common skin conditions (Quilter et al. 2024) into the WHO Skin NTDs app (mainly built as a repository of educational resources and training materials, which adhere to WHO guidelines). While the app aims to improve capacity building through AI, its "real-world" impact on disease management is still not available and upcoming studies will determine its utility. The performance of AI-algorithms is also limited by the availability of images datasets. In the AI4Leprosy study conducted in 2022 at the Brazil leprosy national referral center, although the convolutional neural networks (CNN)-based AI algorithm could contribute to the diagnosis of leprosy with high classification accuracy (90%) (AI4Leprosy), to the best of our knowledge, this has not been validated in a low-resource setting with lack of highly specialized staff. Further afield, while the technology's performance is increasingly being validated in dermatological conditions such as melanoma and other skin cancers (Patel et al. 2023), the direct evidence for its efficacy in diagnosing skin-NTDs remains limited. Overall, while these studies and initiatives demonstrate the potential of AI-powered diagnostic tools, evidence specific to their use for skin NTDs in low-resource settings is still emerging. Our proposed validation study aims to address the aforementioned gaps by evaluating the diagnostic accuracy of an AI-powered app in diagnosing skin NTDs in Kenya, Ethiopia and Senegal, providing critical insights into its practical utility and impact on clinical practice in these settings. The intervention is the SkincAIr Research App, a unified mobile platform containing three role-specific modules: a Dermatologist Dataset eCRF for structured image collection by dermatologists across 5 countries; an FHW eCRF for clinical data collection and case documentation across all 3 study phases; and the SkincAIr Detection App - an AI-powered diagnostic decision-support feature embedded within the FHW eCRF, activated exclusively during Phase B (6 months), and withdrawn during Phase C to assess retention of improved diagnostic skills. The study measures: (SO1) diagnostic performance of FHWs with and without the app, including early detection rate, diagnostic accuracy, sensitivity and specificity against a dermatologist reference standard; (SO2) dataset quality including number, geographic diversity and image quality; (SO3) reduction in diagnostic delay; (SO4) epidemiological surveillance indicators including DHIS2 integration, case confirmation ratio and hotspot identification; (SO5) FHW knowledge gain and user satisfaction. Cost-effectiveness is assessed through primary vs secondary care cost comparison and ICER calculation. This clinical study will provide vital data to assess the practical utility of AI in improving diagnostic accuracy and speed among non-specialist health workers. By leveraging AI and mobile technology, we can equip FHWs in low-resource settings with tools to quickly detect and manage skin NTDs. Justification for the Study: Neglected tropical diseases (NTDs) affect over one billion people globally, with skin NTDs such as leprosy, cutaneous leishmaniasis, and onchocerciasis contributing significantly to morbidity, disability, and stigma in affected populations. Early detection and treatment are crucial to prevent complications, reduce transmission, and improve patient outcomes. In resource-limited settings like Kenya, Ethiopia, Senegal, Nigeria and the Democratic Republic of the Congo, FHWs are often the first point of contact for patients with skin conditions. However, FHWs typically lack specialized training in dermatology, leading to misdiagnosis or delayed diagnosis of skin NTDs. Mobile health (mHealth) interventions, particularly those incorporating AI, offer a promising solution to enhance the diagnostic capabilities of FHWs. The SkincAIr project aims to evaluate whether introducing a mobile app with AI functionality can improve the diagnostic accuracy of FHWs in detecting skin NTDs. Additionally, the study seeks to determine if FHWs retain improved diagnostic skills after the AI assistance is removed, indicating potential capacity building and sustained improvement in healthcare delivery. Our project transcends the twin limitations of lack of specialized training in dermatology and availability of images datasets by collecting real clinical data from multiple geographic locations within Low-and Middle-Income Countries (LMICs), ensuring that our models are trained and externally validated on diverse datasets representative of the target populations. Inclusion of Kenya, Ethiopia, Senegal, Nigeria and the Democratic Republic of the Congo will contribute to the diversity of images that are representative of the Skin NTDs endemic in multiple geographic locations in sub-Saharan Africa. As these locations present unique challenges, including limited healthcare infrastructure and varying disease presentations, the proposed study is crucial in bridging this gap. We will implement a controlled image acquisition protocol to standardize data collection, minimizing the risk of shortcut learning - where models might rely on irrelevant features due to spurious correlations (Winkler et al. 2019). By integrating multimodal data, including clinical parameters, and ensuring robustness to missing data, our models will offer more accurate and personalized diagnostic and predictive capabilities. Null Hypothesis: The use of the SkincAIr AI-powered mobile application will not improve the diagnostic accuracy (sensitivity and specificity) of FHWs in detecting skin-NTDs compared to their baseline performance without the app. General Objective: To develop an AI supported diagnostic tool for skin NTDs (SkincAIr) and test its performance and impact among frontline health workers (FHWs) in detecting select skin NTDs. Specific Objectives: 1. To collect image data for skin NTDs and other skin conditions by dermatologists/dermatology officers or other officers trained on skin NTDS to be used in the training and development of the SkincAIr app. 2. To quantify the impact of the SkincAIr mobile application on the diagnostic accuracy (sensitivity and specificity) of frontline health workers (FHWs) in identifying skin-neglected tropical diseases (skin-NTDs) and other skin conditions. 3. To measure the impact of SkincAIr-assisted diagnosis on key clinical and operational metrics: * Time to diagnosis from first visit - How long it takes for a patient to receive a diagnosis, from the time they first present to a healthcare provider * Time from initial presentation to definitive diagnosis - the time from any initial contact (perhaps community level or first clinic) to the final confirmed diagnosis by a specialist * Reduction in treatment initiation delays * Changes in referral patterns to specialized centers - Changes in how often and how appropriately patients are referred to specialists or higher centers * Cost savings in secondary care attributable to enhanced primary diagnosis (early and accurate diagnoses by FHWs) * New hotspots of suspected or confirmed skin NTDs during the project period (up to M60). 4. To evaluate the sustainability of improved diagnostic skills among FHWs following the withdrawal of AI assistance.