Classification of blue nevus from other lesions for Internet-based melanoma diagnostic system

抄録

Although early detection of malignant melanoma is very important, classification between benign lesion and melanoma is often difficult even by expert dermatologists. Blue nevus often seen as a common benign lesion sometimes has color similar to that of blue white structures, which characterizes malignancy of certain kinds of melanoma. Experienced dermatologists diagnose blue nevus and melanoma relatively easily, however it was still difficult by the computer-based diagnosis. In this paper, we parameterized several clinical items of dermatologist in diagnosing malignant melanoma and built a blue nevus classifier to make our Internet-based melanoma diagnostic system more practicable. A total of 418 dermoscopy images were used in this study and our artificial neural network classifier achieved 76.0% in sensitivity and 99.0% in specificity under the cross-validation test. Over 3/4 of blue nevi were correctly identified without almost no false classification of other nevi or melanomas.