Classification of papulo-squamous skin diseases using image analysis

Abstract

Introduction: Papulo-squamous skin diseases are variable but are very close in their clinical features. They present with the same lesions, erythematous scaly lesions. Clinical evaluation of skin lesions is based on common sense and experience of the dermatologist to differentiate features of each disease.

Aim: To evaluate a computer-based image analysis system as a helping tool for classification of commonly encountered diseases.

Materials and methods: The study included 50 selected images from each of psoriasis, lichen planus, atopic dermatitis, seborrheic dermatitis, pityrasis rosea, and pitryasis rubra pilaris with a total of 300 images. The study comprised three main processes peformed on the 300 included images: segmentation, feature extraction followed by classification.

Results: Rough sets recorded the highest percentage of accuracy and sensitivity of segmentation for the six groups of diseases compared with the other three used techniques (topological derivative, K-means clustering, and watershed). Rule-based classifier using the concept of rough sets recorded the best percentage of classification (96.7%) for the six groups of diseases compared with the other six techniques of classification used: K-means clustering, fuzzy c-means clustering, classification and regression tree, rule-based classifier with discretization, and K-nearest neighbor technique.

Conclusion: Rough sets approach proves its superiority for both the segmentation and the classification processes of papulo-squamous skin diseases compared with the other used segmentation and classification techniques.