Diabetic Foot Ulcers Grand Challenge : Third Challenge, DFUC 2022, Held in Conjunction with MICCAI 2022, Singapore, September 22, 2022, Proceedings

The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Preface Organization Contents Quantifying the Effect of Image Similarity on Diabetic Foot Ulcer Classification 1 Introduction 2 Related Work 3 Methodology 3.1 Dataset Description 3.2 Fuzzy Algorithm 3.3 Identification of Similar Images 3.4 Removal of Similar Images as Determined by Similarity Thresholds 4 Image Similarity Analysis 4.1 Train Set Image Similarity 4.2 Test Set Image Similarity 4.3 Train and Test Set Image Similarity 4.4 Inter-class Image Similarity 4.5 Model Training 5 Results and Discussion 5.1 Baseline Results 5.2 Results on the Curated Datasets 6 Conclusion References DFUC2022 Challenge Papers HarDNet-DFUS: Enhancing Backbone and Decoder of HarDNet-MSEG for Diabetic Foot Ulcer Image Segmentation 1 Introduction 2 Method 2.1 HarDNetV2 – Channel Balanced HarDNet 2.2 Decoder 2.3 Model Ensemble 2.4 Loss Function 2.5 Post-processing 3 Experiments 3.1 Settings 3.2 Dataset 3.3 Experiment Results 4 Conclusion and Future Work References OCRNet for Diabetic Foot Ulcer Segmentation Combined with Edge Loss 1 Introduction 2 Methodology 2.1 Datasets 2.2 Network 2.3 Training 2.4 Post-processing 2.5 Evaluation Metrics 3 Experiments 4 Conclusions References On the Optimal Combination of Cross-Entropy and Soft Dice Losses for Lesion Segmentation with Out-of-Distribution Robustness 1 Introduction 2 Methodology 2.1 The Cross-Entropy Loss, the Dice Loss, and Their Combinations 2.2 On the Definition of In-Distribution and Out-of-Distribution Data for Binary Segmentation Problems 2.3 Model Training Details 3 Experimental Results 3.1 Data and Performance Measures 3.2 Performance Evaluation and Discussion 4 Conclusion References Capture the Devil in the Details via Partition-then-Ensemble on Higher Resolution Images 1 Introduction 2 Proposed Method 2.1 Two-Stage Architecture 2.2 Training Algorithm 2.3 Testing Algorithm 3 Training Details 4 Experiments and Results 5 Conclusion References Unconditionally Generated and Pseudo-Labeled Synthetic Images for Diabetic Foot Ulcer Segmentation Dataset Extension 1 Introduction 2 Dataset and Methods 2.1 Diabetic Foot Ulcer Challenge 2022 Dataset 2.2 Semantic Segmentation via Segmentation Models PyTorch 2.3 Image Synthesis via StyleGAN2+ADA 2.4 Experimental Environment 3 Approach 3.1 Pre-processing 3.2 Augmentations 3.3 Baseline Model Training and Ensemble 3.4 Synthetic Image Generation and Dataset Extension 3.5 Extended Model Training and Ensemble 3.6 Inference and Post-processing 4 Results 4.1 Baseline and Extended Segmentation Ensemble Performance 4.2 Synthetic Images and Pseudo-Labels for Dataset Extension 5 Discussion 5.1 Segmentation Performance and Post-processing 5.2 Dataset Extension via Pseudo-Labeled Synthetic Images 5.3 Limitations 6 Conclusion References Post Challenge Papers Diabetic Foot Ulcer Segmentation Using Convolutional and Transformer-Based Models 1 Introduction 2 Background 3 Related Work 4 Dataset 5 Proposed Method 5.1 Models 5.2 Training Procedure 5.3 Testing Procedure 5.4 Ensemble Models 6 Results 7 Conclusion References Refined Mixup Augmentation for Diabetic Foot Ulcer Segmentation 1 Introduction 2 Dataset 3 Methodology 3.1 Preprocessing 3.2 Proposed Approach 4 Experimental Results 4.1 Fault-Case Analysis 5 Conclusions References DFU-Ens: End-to-End Diabetic Foot Ulcer Segmentation Framework with Vision Transformer Based Detection 1 Introduction 1.1 Related Work 2 DFU-Ens: End-to-End Ensemble DFU Segmentation Framework 2.1 Dataset: Diabetic Foot Ulcer Segmentation Challenge 2022 2.2 Module I: End-to-End Segmentation Framework Based on the U-Net Architecture 2.3 Hybrid Solution Combining Detection and Patch Segmentation with YOLOv4 and Vision Transformers 3 Experimental Results and Analysis 3.1 Hardware and Software 4 Conclusion References Summary Paper Diabetic Foot Ulcer Grand Challenge 2022 Summary 1 Introduction 2 Methodology 2.1 Datasets and Ground Truth 2.2 Performance Metrics 2.3 Summary of the Proposed Methods 3 Results and Discussion 4 Conclusion References Author Index This book constitutes the Third Diabetic Foot Ulcers Grand Challenge, DFUC 2022, which was held on September 2022, in conjunction with the 25th International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI 2022 in Singapore. The 8 full papers presented together with 5 challenge papers and 3 post-challenge papers included in this book were carefully reviewed and selected from 19 submissions. The DFU challenges aim to motivate the health care domain to share datasets, participate in ground truth annotation, and enable data-innovation in computer algorithm development. In the longer term, it will lead to improved patient care.