Multi-Colour Space Channel Selection for Improved Chronic Wound Segmentation

This study introduces a novel approach to chronic wound segmentation, a critical aspect of automated wound monitoring that has the potential to significantly reduce clinical workload. Addressing the challenges posed by varying wound sizes and compositions, our experiments utilise the U-Net architecture with an innovative integration of multiple colour spaces RGB, HSV, CIE-LAB, and YCbCr. Our method involves the merging of various combinations of colour channels from these selected colour spaces. We trained and evaluated our method on the Diabetic Foot Ulcer Challenge 2022 dataset, with improved Intersection-Over-Union (+0.0187), and Dice Similarity Coefficient (+0.0183), in comparison with the baseline model. Additionally, improvements are observed on alternative test sets that include; Complex Wound DB, Advancing the Zenith of Healthcare, and Foot Ulcer Segmentation Challenge datasets. These findings highlight the importance of strategic colour channel selection in chronic wound analysis, and offer a promising direction for future research in medical image analysis. These enhancements show our method’s effectiveness in capturing complex wound characteristics using colour channel selection, contributing a new research direction for medical image analysis.