Dermatological tracking of chronic acne treatment effectiveness

Acne can lead to severe physical and psychological implications on chronic sufferers if not treated promptly and properly. Ramli et al. proposed a k-means cluster based algorithm to provide computer-assisted support for the manual grading of digital images. We propose an improved, automated, and more objective, grading method which involves optimizing the k-means clustering algorithm by identifying the actual number of clusters rather than basing analysis on a fixed K= 3 assumption for all images. The Hough transform was used to further analyze the found acne cluster leading to an approach to more accurately automatically determine the number and type of lesions. A quantitative comparison of the two approaches showed that the new approach provided a better match to the stated specialist analysis. We found it inappropriate to use accuracy and specificity performance analysis metrics to compare the algorithms. A better matching of the algorithms' accuracy to the specialist's analysis of the skin condition was found by modifying the sensitivity metric to account for the Michelson acne grading scale. This robustness suggests that the tool might be a first-step towards patient self-monitoring between visits to a specialist; potentially reducing visits frequency, decreasing wait times, and lead to a definitive standardized assessment scale.