Start Date
9-10-2025 12:00 PM
End Date
9-10-2025 1:30 PM
Description
Objective: To describe the challenges and lessons learned in adults with diabetes mellitus during the image collection phase of the Foot Selfie project, designed to develop and train an artificial intelligence (AI) algorithm for early detection of diabetic foot risk lesions.
Methods: Cross-sectional, mixed-methods study conducted in 2024 in five hospitals in Lima, Peru. Adults with type 1 or type 2 diabetes were recruited using non-probabilistic sampling. Plantar foot images labeled into eight categories were obtained in hospital settings, and a subgroup of participants were followed at home using a mobile application for daily photo capture; field reports were analyzed to identify operational barriers and facilitators.
Results: A total of 812 participants were enrolled (mean age: 60.7 ±10.2 years; 56% female), and 14,685 images were captured, 33% during home follow-up. Most participants were mestizo (71.8%) and had internet access (76%). Prior history of foot ulcers was reported by 33.5% of participants, partial amputations in 15.6% and toe deformities in 46.8%. Within the home follow-up group (n=92), only 3.3% completed the full 30-day period of daily photo capture. Main barriers included technical limitations (incompatible devices, low image resolution), inadequate lighting and low adherence due to competing responsibilities or limited digital literacy. Key facilitators included proactive follow-up by health personnel, caregiver support and educational reinforcement.
Conclusion: Large-scale image collection for AI training is feasible but requires tailored strategies to ensure technical quality, adherence and equity in implementation.
Keywords: diabetic foot, artificial intelligence, early detection, mHealth, lessons learned
Included in
Endocrine System Diseases Commons, Health Information Technology Commons, Public Health Commons
Challenges and Lessons Learned in Image Collection for the Development of an Artificial Intelligence-Based Algorithm for the Early Detection of Diabetic Foot Ulcers
Objective: To describe the challenges and lessons learned in adults with diabetes mellitus during the image collection phase of the Foot Selfie project, designed to develop and train an artificial intelligence (AI) algorithm for early detection of diabetic foot risk lesions.
Methods: Cross-sectional, mixed-methods study conducted in 2024 in five hospitals in Lima, Peru. Adults with type 1 or type 2 diabetes were recruited using non-probabilistic sampling. Plantar foot images labeled into eight categories were obtained in hospital settings, and a subgroup of participants were followed at home using a mobile application for daily photo capture; field reports were analyzed to identify operational barriers and facilitators.
Results: A total of 812 participants were enrolled (mean age: 60.7 ±10.2 years; 56% female), and 14,685 images were captured, 33% during home follow-up. Most participants were mestizo (71.8%) and had internet access (76%). Prior history of foot ulcers was reported by 33.5% of participants, partial amputations in 15.6% and toe deformities in 46.8%. Within the home follow-up group (n=92), only 3.3% completed the full 30-day period of daily photo capture. Main barriers included technical limitations (incompatible devices, low image resolution), inadequate lighting and low adherence due to competing responsibilities or limited digital literacy. Key facilitators included proactive follow-up by health personnel, caregiver support and educational reinforcement.
Conclusion: Large-scale image collection for AI training is feasible but requires tailored strategies to ensure technical quality, adherence and equity in implementation.
Keywords: diabetic foot, artificial intelligence, early detection, mHealth, lessons learned
