Precise Segmentation and Classification of Epithelial Rete-Pegs Signature in Assessing Lower Limb Wound Healing Progression

Abstract

Automated identification of healing progression in lower limb wounds is an important clinical challenge since such wounds fail to heal in an orderly and timely manner. Selection of suitable target features and employment of appropriate analytical tool are critical for the precise monitoring of the healing progression through automated diagnostic method. The present study aims at automated detection of healing trend based on characteristics of epithelial rete pegs, which is considered as an important parameter for monitoring and hence healing progression of epithelial maturation. To this aim, the applicability of various segmentation methods namely, watershed, marker-controlled watershed, k-means clustering and active contour method and comprehensive characterization of rete-pegs are studied and compared. Among them active contour provided better performance in terms of non-linear objective assessments including performance metrics like-peak-signal-to-noise-ratio, processing speed and Mean Square Error. Further, support vector machine is trained for classification of the rete-peg features at different time points of clinically diagnosed non-healing wounds treated with a specific protocol. Kappa scoring tests inter-rater agreement between applied segmentation protocol and ground truth images. The overall accuracy of the system has been found to be 95%. It is thus concluded that with the aid of the proposed segmentation and classification method an automated detection system can be developed for precise monitoring of lower limb wound healing progression.