Automatic Clustering-Based Segmentation and Plaque Localization in Psoriasis Digital Images

Abstract

Psoriasis is one of the most stressful skin diseases. The accurate assessment and effective management of the disease is one of the contributing factors in reducing the time required for relieving the disease symptoms. As the treatment is unusually subjective, an automatic and efficient computer aided assessment technique is an active area of research. In this study, we developed an automatic psoriasis segmentation and plaque localization system using images captured by a digital camera. Our work differs from other studies by improving the segmentation of lesion regions and using a novel region based color feature extractor for classification of psoriasis plaques from healthy skin areas. The proposed modified k-means clustering based segmentation approach resulted in an accuracy of 93.83% in comparison to the ground truth, which is around 10% more than reported results by others with the same database. Statistical analysis was performed to determine psoriasis biomarkers, and the effectiveness of these biomarkers was validated by developing a machine learning model consisting of support vector machine (SVM) classifier to identify the psoriasis plaques automatically. The classification model predicted the disease plaques with an acceptable accuracy of 86.83%, and thus the automated psoriasis segmentation and plaque localization technique developed in this study provide the foundation towards designing an objective assessment system for psoriasis.