Deep Learning in Diabetic Foot Ulcers Detection: A Comprehensive\n Evaluation

Abstract

There has been a substantial amount of research involving computer methods\nand technology for the detection and recognition of diabetic foot ulcers\n(DFUs), but there is a lack of systematic comparisons of state-of-the-art deep\nlearning object detection frameworks applied to this problem. DFUC2020 provided\nparticipants with a comprehensive dataset consisting of 2,000 images for\ntraining and 2,000 images for testing. This paper summarises the results of\nDFUC2020 by comparing the deep learning-based algorithms proposed by the\nwinning teams: Faster R-CNN, three variants of Faster R-CNN and an ensemble\nmethod; YOLOv3; YOLOv5; EfficientDet; and a new Cascade Attention Network. For\neach deep learning method, we provide a detailed description of model\narchitecture, parameter settings for training and additional stages including\npre-processing, data augmentation and post-processing. We provide a\ncomprehensive evaluation for each method. All the methods required a data\naugmentation stage to increase the number of images available for training and\na post-processing stage to remove false positives. The best performance was\nobtained from Deformable Convolution, a variant of Faster R-CNN, with a mean\naverage precision (mAP) of 0.6940 and an F1-Score of 0.7434. Finally, we\ndemonstrate that the ensemble method based on different deep learning methods\ncan enhanced the F1-Score but not the mAP.\n