Artificial intelligence for the diagnosis and prognosis of leprosy, leishmaniasis, and schistosomiasis: A systematic review

Background: Neglected Tropical Diseases affect over one billion individuals globally, challenging healthcare systems in endemic regions. While promising, Artificial Intelligence (AI) requires rigorous validation for clinical translation. Methodology: This systematic review of prediction-model studies evaluated AI applications for leprosy, schistosomiasis, and leishmaniasis (2015–2025). Methodological quality and reporting transparency of the included studies were appraised using the PROBAST+AI and TRIPOD+AI tools, respectively. Findings: Among 69 studies identified, 89.9% were published after 2020. The literature predominantly focused on diagnostic support (81.2%), leaving the prognosis of adverse outcomes underexplored (18.8%). Methodological fragility was widely observed: 89.9% of models exhibited a high risk of bias, primarily due to lack of external validation (82.6%) and failure to report or appropriately handle missing data (66.7%). Transparency remains a barrier, with source code unavailable in 76.8% of cases, hindering reproducibility. Conclusion: Current evidence indicates that quantitative output outpaces methodological maturity. Bridging the gap between algorithmic potential and public health impact requires prioritizing open science and robust validation. Author summary: Leprosy, leishmaniasis, and schistosomiasis continue to pose significant challenges to global health, disproportionately affecting communities in resource-limited regions. Artificial Intelligence (AI) has emerged as a promising technology to improve how we identify and monitor these diseases. However, the move from laboratory development to clinical practice requires a high level of scientific rigor to ensure that these tools are safe and effective for patient care. By reviewing 69 studies, we found that the recent surge in AI research has not been matched by methodological maturity. Our analysis shows that 89.9% of the developed models were classified as having a high risk of bias. This vulnerability stems from common practices such as using data from only one hospital or failing to test the AI on patients from different geographic areas. Furthermore, scientific transparency remains low, as most researchers do not share the underlying code needed to verify their results. Our study highlights that the current focus on producing more models must be replaced by a commitment to robust validation and open science. For AI to become a reliable tool for clinicians in endemic regions, future research must address these technical shortcomings. Without these changes, AI risks remaining an experimental concept rather than a practical solution for disease control.