Diabetic Foot Ulcer Grand Challenge 2021: Evaluation and Summary

Diabetic foot ulcer classification systems use the presence of wound infection (bacteria present within wound) and ischaemia (restricted blood supply) as vital clinical indicators for treatment prediction healing. Studies investigating automated computerised methods classifying diabetic wounds are limited due to a paucity publicly available datasets severe data imbalance in those few that exist. The Foot Ulcer Challenge 2021 provided participants with more substantial dataset comprising total 15,683 patches, 5,955 used training, 5,734 testing an additional 3,994 unlabelled patches promote development semi-supervised weakly-supervised deep learning techniques. This paper provides evaluation 2021, summarises results obtained from each network. best performing network was ensemble top 3 models, macro-average F1-score 0.6307.