Translating Clinical Delineation of Diabetic Foot Ulcers into Machine Interpretable Segmentation

—Diabetic foot ulcer is a severe condition that requires close monitoring and management. For training machine learning methods to auto-delineate the ulcer, clinical staff must provide ground truth annotations. In this paper, we propose a new diabetic foot ulcer dataset, namely DFUC2022, the largest segmen- tation dataset where ulcer regions were manually delineated by clinicians. We assess whether the clinical delineations are machine interpretable by deep learning networks or if image processing refined contour should be used. By providing benchmark results using a selection of popular deep learning algorithms, we draw new insights into the limitations of DFU wound delineation and report on the associated issues. With in depth understanding and observation on baseline models, we propose a new strategy for training and modify the FCN32 VGG network to address the issues. We achieved notable improvement with a Dice score of 0.7446, when compared to the best baseline network of 0.5708 and the first place in DFUC2022 challenge leaderboard, with a Dice score of 0.7287. This paper demonstrates that image processing using refined contour as ground truth can provide better agreement with machine predicted results. Furthermore, we propose a new strategy to address the limitations of the existing training protocol. For reproducibility, all source code will be made available upon acceptance of this paper, and the dataset is available upon request.