Deep learning in chronic wound segmentation : a comprehensive review and meta-analysis

Cassidy, Bill and Kendrick, Connah and Reeves, Neil D. and Pappachan, Joseph M. and Yap, Moi Hoon (2025) Deep learning in chronic wound segmentation : a comprehensive review and meta-analysis. Visual Computer, 41 (14). pp. 11885-11908. ISSN 0178-2789

Text (371_2025_Article_4133.pdf)
371_2025_Article_4133.pdf - Published Version
Available under License Creative Commons Attribution.
Download (2MB)

Abstract

This work conducts a review of all chronic wound segmentation deep learning studies meeting specific criteria that have been published since research first started in this domain 8 years ago (2015–2023). Management of chronic wounds represents a serious ongoing concern for hospitals and outpatient clinics world-wide. There is a clear need for technological interventions using deep learning approaches that could have a potential significant impact in the automated monitoring of such wounds. We review the existing literature and perform R-squared statistical analysis to form a fresh understanding of the field to gain deeper insights into the issues that are presenting obstacles to research progress. Our findings show a negative correlation between small test set size and test metrics (Dice similarity coefficient and mean intersection over union), indicating smaller test sets are associated with higher test metrics. We also identify other major hurdles in the field, such as a lack of data understanding, a lack of data availability, and a lack of research transparency. The focus of this body of work is to increase understanding of the underlying issues that have pervaded in deep learning chronic wound research. A clear presentation of findings in this work can be used by researchers as a guide to avoiding common pitfalls, and to advance research knowledge.