Utilizing Mask R-CNN for Automated Evaluation of Diabetic Foot Ulcer Healing Trajectories: A Novel Approach

The healing trajectory of diabetic foot ulcers (DFUs) is commonly determined through manual inspection, a method which is often subjective and prone to errors. In an effort to address these limitations, this study explores an artificial intelligence-based, computer-aided assessment technique as an alternative. This approach leverages the power of a Hue, Saturation, and Value (HSV)-based image fusion technique, integrating thermal and visual data to deliver precise wound characterizations. Further, through the deployment of an instance segmentation-based Mask Region-based Convolutional Neural Network (Mask-RCNN), the area of the wound is estimated. This randomized, prospective, single-blind study was conducted over a 12-week period, focusing on neuropathic DFUs (Wagner grade 2) located on the plantar aspect of the foot. Forty-two patients were enrolled, with an average age of 54.28 +/- 7.45 years and an average ulcer duration of 5.86 +/- 2.22 years. The healing trajectory of eight patients, observed weekly, was further analyzed. The absolute temperature difference (ATD) between contralateral ulcer regions was found to be 2.63 +/- 1.99 degrees C, with the respective z-score values of ATD providing a significant p-value of 0.000040412 (i.e., p<0.05). The correlation between the ground truth (ulcer area estimation by clinicians using Woundly software) and the proposed method was found to be at an average of 92.50%. The study ultimately concludes that the Mask-RCNN technique, when applied to fused images, can facilitate automated and user-independent assessments of DFUs. This method has the potential to aid in the accurate characterization of healing trajectories, thereby enhancing the overall understanding of wound progression in diabetic patients.