AI Tools for Leprosy

94 published methods that apply artificial intelligence to leprosy, from 1994 to 2026, placed against the WHO target product profiles by intended use and against the WHO stages for evaluating AI-based medical devices.

Skin Lesion Images, Leprosy Only

• Deep Learning Detector for Hypopigmented Skin Disease, 2026. Deep learning. DOI
• Synthetic Leprosy Images by Latent Diffusion, 2026. DeepLabV3 segmentation and mask-conditioned Stable Diffusion, transferred from wound images. A set of synthetic images for training other models. arXiv 2609.13226
• Convolutional Neural Network and Support Vector Machine Compared (ETCOM 2025), 2025. Convolutional neural network and support vector machine. DOI
• Deep Learning Detection of Leprosy Lesions (ICCCNT 2024), 2024. Deep learning. DOI
• Lesion Segmentation with Deep Spatial Features, 2024. Segmentation, convolutional network features and a machine learning classifier. DOI
• Semantic Segmentation with Spatial and Texture Features, 2024. Semantic segmentation with convolutional network and texture features. DOI
• Few-Shot Siamese Network for Lesion Detection, 2023. Siamese network trained on few examples. 91% in training and 73% in testing. leprosy-information.org resource
• Intelligent System for Skin Cancer and Leprosy (UFPE), 2023. Deep neural networks and random forest on digital images. LA Referencia 2023 · DOI
• LeprosyNet, an Explainable Convolutional Neural Network, 2023. Convolutional neural network with activation maps and occlusion sensitivity. DOI
• AI4Leprosy (Fiocruz, Microsoft and Novartis Foundation), 2022. Convolutional neural networks on images, and elastic-net logistic regression on images with clinical data. About 90% accuracy for the combined model. PubMed 36776278
• Early Segmentation and Deep Learning for Leprosy Patches, 2017. Image segmentation and deep learning. DOI
• Lesion Recognition by Convolutional Neural Network (Baweja et al.), 2016. Convolutional neural network. 91.6% accuracy. DOI
• CMATER-JU Classifiers for Hypopigmented Lesions, 2013. Texture features with a support vector machine (2013), a feature ensemble (2020 and 2023), and GAN augmentation with DenseNet-121 (2020). 94.25% accuracy with GAN augmentation and DenseNet-121. DOI · arXiv 2004.04122 · DOI

Skin Images, Several Diseases

• Skin NTD Image and Clinical Data Models (Arba Minch, Ethiopia), 2026. Features from pretrained networks, a custom network, and eight models on tabular data. PubMed 42337620 · PubMed 42628011
• ResNet-50 Skin Disease Classifier (INCET 2025), 2025. ResNet-50. 94.8% validation accuracy. DOI
• eSkinHealth Dataset and Concept Models, 2025. Dataset annotated with language and segmentation models under dermatologists' guidance. A benchmark for other models. arXiv 2508.18608 · DOI
• WHO Skin NTDs App (UniversalDoctor), 2024. DenseNet-121 image classifier with a decision support tool from NLR. Top-five sensitivity 99.8% in internal testing. In an independent test, leprosy was among the top five suggestions for 84.9% of images, and 69.6% for reactional and atypical lesions. PubMed 42102907 · PubMed 42006576 · PubMed 40971863 · PubMed 42447456 · Leprosy Review
• YOLOv9 Skin Lesion Detector, 2024. YOLOv9. Mean average precision 81.4%. DOI
• Deep Learning for Skin NTDs in Côte d'Ivoire and Ghana (Yotsu et al.), 2023. ResNet-50 and VGG-16. More than 70% of diagnoses correct. PubMed 37578966 · medRxiv
• Neural Network Tuned by Evolutionary Algorithms on Raspberry Pi, 2023. Artificial neural network optimised by evolutionary algorithms. DOI
• Real-Time Recogniser of Skin NTD Lesions, 2022. Feature extraction and a support vector machine. Sensitivity 92%, specificity 94%, accuracy 96%. DOI
• Deep Facial Diagnosis, 2020. Transfer learning from face recognition networks. Top-one accuracy above 90%. DOI
• Recognisers for Common Skin Diseases (Philippines), 2020. Support vector machine and neural network (2020), and YOLOv5 (2024). About 93.5% precision for the support vector machine. DOI · DOI

Microscopy and Histopathology

• Histopathology Classifiers for Leprosy (H&E), 2025. Neural network, decision tree, logistic regression and random forest. Neural network sensitivity 74%, specificity 65%, AUC 0.71. PubMed 39498903
• No-Code Classifiers of Acid-Fast Bacilli (Tuberculosis and Leprosy), 2024. Three no-code image classification platforms. For leprosy, sensitivity 100% and specificity 96% with the best platform. That model kept training on images during the test period. PubMed 38389642
• Mask R-CNN Detector of Acid-Fast Bacilli in Slit-Skin Smears, 2023. Mask R-CNN (ResNet-50) on Raspberry Pi. DOI

Nerve Function and Imaging

• Digitiser for Simplified Neurological Assessment Forms, 2025. YOLOv8x, YOLO11x and Faster R-CNN. Precision above 94%. DOI · DOI
• Nerve Segmentation in Ultrasound Videos, 2025. YOLOv8 with X-Mem video segmentation. A first baseline for this task. DOI
• Facial Surface Electromyography Classifier (UFU), 2016. Support vector machine. DOI
• SeyeS Support System for Ocular Disability, 2010. Bayesian network. Sensitivity 60%, specificity 100%. DOI
• Electromyography Expert System, 2004. Wavelet features, fuzzy rules and a back-propagation neural network. DOI
• Discriminant Classifier of Nerve Conduction, 1995. Discriminant function. Classed clinically unaffected nerves as abnormal. PubMed 7594923

Ulcers and Chronic Wounds

• Image-Processing Framework for Leprosy Wounds (Mehta et al.), 2026. Image enhancement, UNet++ segmentation and a hybrid of VGG-19 and Inception. PubMed 41553957 · DOI
• LUTSeg Dataset and TiSage Segmentation, 2026. Semi-supervised teacher and student model with priors from a medical vision and language model. Better than baseline methods when few labels are available. arXiv 2608.25866
• AI-Enabled Multispectral Imaging Device for Leg Ulcers, 2025. Point-of-care autofluorescence imaging with AI, a commercial device. Wound area fell by 84% with the device and by 73% with standard care. PubMed 40981688
• YOLO and U-Net Detection and Segmentation of Wounds, 2025. YOLO models and U-Net. DOI
• Ulcer Care App (Nepal), 2024. Models predicting ulcer outcome from photographs and clinical data. DOI
• Smartphone Tracking of Chronic Wounds and the CO2Wounds Dataset, 2023. Deep learning detection and segmentation. Precision up to 84.5%. PubMed 37633087 · arXiv 2408.10827

Host Biomarkers and Molecular Data

• FTIR Spectroscopy of Saliva for Leprosy Reactions, 2026. Orthogonal signal correction and partial least squares discriminant analysis. Sensitivity 100%, specificity 100%. DOI
• Classifier of Antimicrobial Proteins in Reversal Reactions, 2025. Classifier for membrane-permeabilising proteins. 28 genes predicted, four confirmed active against M. leprae. PubMed 40569678
• Decision Trees on Plasma Immune Mediators (Carvalho et al.), 2025. Decision trees on a multiplex panel. AUC up to 0.98 for CXCL8 in one comparison. PubMed 40109344
• Three-Gene Signature Including Non-Coding RNAs, 2025. Machine learning on RNA sequencing, including non-coding genes (SNHG5, SNHG8, C6orf48). Sensitivity 88%, specificity 88%, AUC 0.96. PubMed 41455820
• Decision Tree for a Cytokine Release Assay (Marçal et al.), 2022. Decision tree. Correct in 79–84% of contacts and 70–73% of patients. PubMed 35169594
• Algorithm for Susceptibility and Clinical Course (UFJF), 2021. Random forest on cytokines and genetic data. DOI
• RISK4LEP Blood RNA Signature, 2021. Random forest for feature selection. AUC 0.86, predicting leprosy 4–61 months before diagnosis. PubMed 34090257
• Random Forest Combining Serology and qPCR (Gama et al.), 2019. Random forest. Sensitivity 90.5% in multibacillary and 70.6% in paucibacillary leprosy, specificity 92.5%. PubMed 31181059
• Classification Tree for Multiplex Pseudogene PCR, 2017. Classification and regression tree. Predicted positivity in groups with negative RLEP or a low bacterial index. PubMed 28317797
• Bayesian and Neural Networks on Genetic Association Data, 2016. Bayesian network, neural network, logistic regression and splines. The Bayesian network performed best. PubMed 27538955
• Neural Network Selection of M. leprae-Specific Peptides, 2012. Artificial neural network. A subset of peptides followed the intensity of transmission. PubMed 22545169
• Decision-Tree Cut-Offs for qPCR (sodA, 16S rRNA, RLEP and Ag85B), 2011. Decision trees. RLEP sensitivity 87.1%. PubMed 22022631

Clinical and Routine Health Data

• Machine Learning Framework for Leprosy Reactions (Yunnan), 2026. Nine models, with naive Bayes and XGBoost explained by SHAP. Naive Bayes test AUC 0.76, sensitivity 44%, specificity 92%. DOI
• Prediction of Disability Grade, Trees and Neural Networks Compared, 2026. Random forest, LightGBM, CatBoost, multilayer perceptron, ResNet and transformer. LightGBM AUC 0.71. DOI
• Risk of Misdiagnosis (Yunnan), 2026. Nine models, with LASSO, SMOTE and SHAP. Neural network test AUC 0.68, sensitivity 78%, specificity 57%. PubMed 42180479
• Disability Prediction in a 39-Year Cohort (Wenshan), 2025. Nine models, with random forest best, explained by SHAP. AUC 0.92 in internal validation and 0.70 on the external test set. DOI
• MaLeSQs (Machine Learning Leprosy Suspicion Questionnaire), 2025. Support vector machine, compared with logistic regression, random forest and XGBoost. Sensitivity 85.7%, specificity 69.2%. Sensitivity 100% when combined with serology. PubMed 40011614 · PubMed 41501696 · Research Square preprint rs-4402230
• Predictors of Late Diagnosis with Grade 2 Disability (Brazil), 2025. Random forest, LightGBM, CatBoost, XGBoost and an ensemble. AUC 0.93. PubMed 40423361
• Bayesian Network for the Risk of Leprosy Reactions (Online Tool), 2023. Bayesian network built with WEKA and Netica. Sensitivity 79.3%, specificity 86.2%, accuracy 82.7% (87.7% with genetic data). PubMed 37564037
• Classification Tree of Persuading Factors for Diagnosis, 2021. Classification and regression tree. DOI
• Cross-Platform Screening App (SINAN Data), 2021. Random forest. Sensitivity 93.97%, specificity 87.09% for operational classification. PubMed 33825685
• Text Mining and Severity Classifiers for Health Records (India), 2020. Language processing with rule-based, support vector machine and logistic regression models (2020), AdaBoost and XGBoost (2023), and an LSTM network. 97% accuracy for the ensemble. DOI · DOI
• Classification Tree for Disability at Release from Treatment, 2019. Classification and regression tree. A clinical score. PubMed 30785567
• Neural Network Prediction of Leprosy from Demographic and Clinical Data, 2017. Artificial neural network. Semantic Scholar
• CHAID Decision Trees for Social Functioning and Progression of Disability, 2009. CHAID decision trees. PubMed 22783757 · PubMed 19287872

Expert Systems, Decision Support and Language Models

• ChatLinoHans Chatbot for Lesion Management, 2026. Chatbot with validated content. The abstract does not say whether it uses a language model. Content validity index 0.98, usability score 77.5–100. PubMed 42385062
• SkinSelfcareQA Dataset (Nigeria), 2026. Question and answer dataset for language models, created with affected people. DOI
• Treatment Recommender for Tropical Diseases (Indonesia), 2026. Bayesian network and association rules. DOI
• Generative AI Chatbot for Nursing Education on Leprosy, 2025. Generative AI, from a scoping review to a chatbot. OSF Preprints 2025
• Language Models on Dermatology, Venereology and Leprosy Questions, 2025. ChatGPT 3.5 and Bing Chat. Bing Chat 78%, ChatGPT 60% and postgraduates 43% correct. Both models were less accurate than the postgraduates on inflammatory disorders and leprosy. PubMed 40125046
• Health Guardian (Brazil), 2024. App with AI support for treatment decisions. DOI
• Fuzzy Classifier for Diagnosis and Prognosis, 2012. Fuzzy classifier. DOI
• Text Analysis of Psychosocial Interviews, 1998. AI text analysis. PubMed 10188555
• LEPDIAG Fuzzy Expert System, 1994. Fuzzy production rules (FRUIT) with a homeostatic expert module. DOI · Sādhanā, 1994 · Information Sciences, 1995

Epidemiology and Surveillance

• Machine Learning Projections of Leprosy in Brazil to 2034, 2026. Random forest and statistical forecasting. Elimination not projected by 2030. PubMed 42066748
• Cluster Detection for Post-Exposure Prophylaxis (India), 2022. DBSCAN clustering with expert rules. 75% and 57% of cases captured. PubMed 36525390
• Explainable Models of Treatment Patterns (Piauí), 2022. Regression trees. DOI
• Bayesian Network Analysis of Socio-Demographic Factors (Legal Amazon), 2019. Bayesian networks. LA Referencia 2019
• Vulnerability Index from Spatial Analysis and AI (Amazon), 2019. AI methods with spatial analysis. DOI
• Self-Organising Maps for Leprosy Epidemiology, 2018. Kohonen self-organising maps. DOI
• Fuzzy Rule-Based Incidence Model (Cuiabá), 2015. Fuzzy rule-based system. DOI
• Web Page Classifier for Tropical Disease Reports (Indonesia), 2011. Support vector machine, naive Bayes and k-nearest neighbours. F-score 95.5%. DOI · DOI

Drug Discovery, Resistance and Protein Structure

• HANSEN Structural Proteome, 2026. AlphaFold3, Boltz and Chai-1 structure prediction, with pocket and ligand mapping. A public resource. DOI
• ML0099 Target Discovery (AlphaFold3 and DeepTMHMM), 2026. AI structure prediction with docking and molecular dynamics. Zileuton and glipizide proposed. PubMed 42594574
• Machine Learning QSAR Screen of Plant Compounds against InhA, 2026. Machine learning QSAR with docking and molecular dynamics. Hinokiflavone proposed as a lead. PubMed 41575667
• Pooled AlphaFold3 Map of M. leprae Protein Interactions, 2026. Pooled AlphaFold3 across all of about 1.3 million possible pairs. About 2,000 strong or intermediate interactions, with functions predicted for dozens of uncharacterised essential proteins. DOI
• Random Forest Surrogate for Antileprosy Activity, 2026. QSAR descriptors with a random forest surrogate model. R² 0.87. DOI
• TLR2 Stochastic Model with Deep Learning, 2026. Stochastic differential equations, topological data analysis and neural networks. DOI
• AlphaFold Models for Vaccine Design and HBHA–APOA1 Docking, 2024. AlphaFold structures with docking. DOI · DOI
• AlphaFold Mapping of the FtsQBL Divisome, 2022. AlphaFold, AlphaFold-Multimer and hidden Markov models. Remote homologues assigned. PubMed 36685953
• APRANK Antigen Prioritisation, 2021. Ranking model combining localisation, disorder, repeats and epitope predictions. Ranked lists of antigens. DOI
• HARP and rpoB Saturation Mutagenesis, 2020. mCSM machine learning predictors of stability and affinity. A consensus impact for each mutation. PubMed 33304465 · PMC7000446
• SUSPECT-RIF, 2020. Structure-based machine learning with graph signatures. Sensitivity 92%, specificity 84% in M. tuberculosis. PubMed 33093532
• Membrane Protein Predictors for Mycobacteria (Bi-PSSM and Unb-DPC), 2017. Support vector machine on position-specific and dipeptide features. 95–99.6% accuracy. PubMed 28927812 · PubMed 27939596
• Optimisation of Rifamycin B Fermentation, 2004. Genetic algorithm with a decision tree. Productivity raised by more than 600%. PubMed 15052640

Leprosy in Animals

• PanLep300 Detector of Leprosy in Wild Chimpanzees, 2026. Image and video classifiers (2D, 2.5D and 3D). Combining image-level predictions worked best. arXiv 2607.17395