Deep Learning for Wound Tissue Segmentation: A Comprehensive Evaluation using A Novel Dataset

TL;DR

This paper addresses the lack of publicly accessible wound tissue segmentation data by introducing the WoundTissue dataset, consisting of 147 wound images labeled for six tissue types and prepared in full-image, patch, and superpixel formats. It conducts an extensive benchmark of 82 DL models across segmentation and classification tasks, comparing conventional and modified architectures and evaluating class-weighted versus unweighted training. The study finds that multi-scale architectures like FPN with VGG16 generally deliver the best tissue-segmentation performance, achieving a high Dice score in leave-one-out cross-validation, while underrepresented tissues such as bone and tendon remain challenging. The publicly available dataset and comprehensive evaluation establish a benchmark framework to guide future research and clinical translation in automated wound analysis, with suggested directions including data augmentation, custom losses, and ensemble models to improve performance on difficult tissue types.

Abstract

Deep learning (DL) techniques have emerged as promising solutions for medical wound tissue segmentation. However, a notable limitation in this field is the lack of publicly available labelled datasets and a standardised performance evaluation of state-of-the-art DL models on such datasets. This study addresses this gap by comprehensively evaluating various DL models for wound tissue segmentation using a novel dataset. We have curated a dataset comprising 147 wound images exhibiting six tissue types: slough, granulation, maceration, necrosis, bone, and tendon. The dataset was meticulously labelled for semantic segmentation employing supervised machine learning techniques. Three distinct labelling formats were developed -- full image, patch, and superpixel. Our investigation encompassed a wide array of DL segmentation and classification methodologies, ranging from conventional approaches like UNet, to generative adversarial networks such as cGAN, and modified techniques like FPN+VGG16. Also, we explored DL-based classification methods (e.g., ResNet50) and machine learning-based classification leveraging DL features (e.g., AlexNet+RF). In total, 82 wound tissue segmentation models were derived across the three labelling formats. Our analysis yielded several notable findings, including identifying optimal DL models for each labelling format based on weighted average Dice or F1 scores. Notably, FPN+VGG16 emerged as the top-performing DL model for wound tissue segmentation, achieving a dice score of 82.25%. This study provides a valuable benchmark for evaluating wound image segmentation and classification models, offering insights to inform future research and clinical practice in wound care. The labelled dataset created in this study is available at https://github.com/akabircs/WoundTissue.

Figures (6)

• Figure 1: Overview of wound segmentation research and the scope of this study
• Figure 2: Different types of wound tissues in the dataset.
• Figure 3: Patching and superpixels
• Figure 4: A taxonomy of different deep learning approaches for wound tissue classification and segmentation.
• Figure 5: Visual comparison of predicted wound tissue segmentation using two best-performing models across three forms, D1, D2, and D3, of the dataset. The first column from the left shows input images, while the second column illustrates the ground truth, where different tissue types are colour-coded: granulation in red, slough in yellow, maceration in white, necrosis in black, tendon in sky blue, and bone in cream. The remaining columns display the predicted segmentation results from the two best-performing models for each dataset form, showcasing the accuracy and limitations of each model in identifying and segmenting the wound tissues.