Prediction of Diabetic Foot Ulcer Healing Upon Initial Visit Using Artificial Intelligence: A Preliminary Retrospective Study

Approximately 15-25% of Americans with diabetes develop a diabetic foot ulcer (DFU) [1,2]. These wounds lead to a loss of mobility and lower quality of life [3]. It is generally accepted that DFUs with greater than 50% area reduction (PAR) after 30 days are likely to heal by 12 weeks [4-7]. However, using this metric requires four weeks of treatment before one can determine if a more effective therapy should be used. An earlier and more accurate means of predicting DFU healing is important to immediately select the best treatment, and reduce time to healing.

We proposed an artificial intelligence algorithm could make an accurate prediction of 30-day PAR with the information obtained during the initial wound assessment. To test this hypothesis, medical history and images from 118 subjects and 150 unique DFUs were gathered retrospectively under an IRB study. From these images, we developed an artificial neural network to reduce the image complexity, concatenate image data with medical history, and predict the probability of 50% PAR at day 30. Algorithm results were obtained using cross-validation.

Image data taken during the initial visit could predict 50% PAR with 67% (± 22 sd) accuracy. Patient medical history alone was 76% (± 18 sd) accurate, with the most important variables being: wound area; BMI; number of previous wounds; HbA1c; kidney failure; type II vs type I diabetes; anemia; asthma; drug use; smoking status; diabetic neuropathy; DVT; and previous MI. When combining these medical variables with image data we observed a modest increase in prediction accuracy to 78% (± 18 sd).

These preliminary results show that an algorithm based on information obtained at the initial patient visit could accurately predict 50% PAR at 30 days. In addition, medical data combined with images of the wound increased prediction accuracy.