Development of a Non-Invasive Imaging Device for Prediction of Diabetic Foot Ulcer Healing Potential

Background: Diabetic foot ulcer (DFU) incidence in the diabetic population has been reported between 15-25% and leads to increased morbidity and mortality, higher risk of amputation, and millions of dollars in healthcare cost annually. Current reported DFU healing rates are approximately 30% by 12 weeks and 45% overall. Standard clinical practice evaluates DFU healing via changes in wound size and appearance after a time of routine wound treatment. This can delay the application of more effective treatment methods. Ability to predict wound healing potential at the initial DFU evaluation could significantly decrease periods of healing stasis and increase the resolution rates of DFUs.

Methods: For this IRB-approved study, multi-spectral imaging data was obtained from 8 DFUs using a non-contact imaging device during the initial clinic visit. This data was analyzed by an artificial intelligence (AI) deep learning program to predict areas of non-healing within the DFUs. After 4 weeks, actual area of non-healing wound bed was compared to AI prediction. Accuracy of the AI program was evaluated using cross-validation due to the small sample size.

Results: The AI program demonstrated a sensitivity of 98.8% and specificity of 93.9% in predicting areas of the DFUs that would not heal by the 28-day healing assessment. Every DFU displayed some degree of healing but none of the DFUs healed completely within 28 days.

Conclusion: Although this initial study sample size is small, the use of AI with multi-spectral imaging data for DFU healing prediction is promising. Further training of the AI program on a larger patient population is currently underway to develop a more robust prediction algorithm.