OCRNet for Diabetic Foot Ulcer Segmentation Combined with Edge Loss

Huahui Yi, Wei Xu, Zekun Jiang, Jun Gao, Qingbo Kang, Qicheng Lao, Kang Li

Abstract: Diabetic foot ulcer is a serious manifestation of lesions on the diabetic foot that requires close monitoring and management. The research at hand investigates an approach on segmentation of diabetic foot ulcer area, conducted as part of the Diabetic Foot Ulcer Challenge (DFUC) 2022. We use OCRNet as the baseline for segmentation and a powerful ConvNeXt network was adopted as the backbone. To obtain better results, a boundary loss was introduced to further constrain the boundary of segmentation. In addition, gamma correction was used in the inference stage in order to reduce the difference in luminance between the training, validation and test sets. Our method won 2nd place in the DFUC2022 with a Dice score of 72.80%. Source code is available at: DFUC2022SegmentationOcrnet .