Spectral Clustering for Unsupervised Segmentation of Lower Extremity Wound Beds Using Optical Images

Area of Science:

• Biomedical Engineering
• Medical Imaging Analysis
• Computational Dermatology

Background:

• Chronic lower extremity wounds affect many elderly individuals globally, posing significant healthcare costs.
• Current wound assessment relies on subjective visual inspection, limiting accuracy and efficiency.
• Accurate wound evaluation is crucial for effective management and monitoring healing progression.

Purpose of the Study:

• To develop and validate a novel spectral clustering method for automated wound-area detection and segmentation.
• To enhance the accuracy and objectivity of chronic wound assessment using digital imaging.
• To provide a robust tool for measuring ulcer perimeters and tracking healing.

Main Methods:

• Utilized digital images captured by a hand-held optical camera for wound analysis.
• Employed a spectral clustering (SC) approach based on an affinity matrix and the Ng-Jorden-Weiss algorithm.
• Pre-processed images with color homogenization and first-order statistics filtering, followed by morphological operators for post-processing.

Main Results:

• The SC method achieved a segmentation accuracy of 86.73%, with positive predictive values of 91.80% and sensitivity of 89.54% on 105 images.
• Demonstrated superior performance compared to k-means and Fuzzy C-Means (FCM) clustering algorithms.
• Validated segmentation accuracy against dermatologist-labeled ground-truth images.

Conclusions:

• The proposed spectral clustering method offers a robust and accurate approach for chronic wound segmentation.
• This tool has the potential for integration into patient-facing systems for rapid clinical assistance and improved wound care.
• The technique supports reliable ulcer perimeter measurement and monitoring of healing progression.