The classification of melanocytic skin lesions using ultrasonic data

Abstract

The purpose of this study was to evaluate parameters for tissue characterization and to develop the algorithm for automatic recognition of malignant melanoma from melanocytic nevi by using ultrasonic data. The shape, acoustical and textural parameters of melanocytic skin lesions were tested. The linear regression classifier was applied in order to differentiate malignant skin tumours from benign. The most informative features were selected using Mahalanobis distance. The averaged area under the ROC curve was 0.840±0.028. The results showed that it is possible to differentiate malignant and benign melanocytic skin lesions by using ultrasonic data. Automatic classification based on ultrasonic data can supplement existing diagnostic methods in clinical dermatology.