Computer assessment of the composition of a generic wound by image processing

Aim: This paper addresses the assessment of the composition of a general wound, in terms of all identifiable categories of tissue and pigmentation in an attempt to improve accuracy in assessing and monitoring wound health. Materials and Methods: A knowledgebase of clusters was built into the hue, saturation and intensity (HSI) color space and then used for assessing wound composition. Based on the observation that the clusters are fairly distinct, two different algorithms, that is, Mahalanobis distance (MD) based and the rotated coordinate system (RCS) method, were used for classification. These methods exploit the shape, spread, and orientation of each cluster. Results: The clusters in the HSI color space, built from about 9,000 (calibrated) pixels from 48 images of various wound beds, showed 8 fairly distinct regions. The inter‐cluster distances were consistent with the visual appearance. The efficacy of the MD and RCS based methods in 120 experiments taken from a set of 15 test images (in terms of average percent‐match) was found to be 91.55 and 93.71, respectively. Conclusion: Our investigations established eight categories of tissue and pigmentation in wound beds. These findings help to determine the stage of wound healing more accurately and comprehensively than typically permitted through use of the 4‐color model reported in the literature for addressing specific wound types.