COMPUTER IMAGE-ANALYSIS IN THE DIAGNOSIS OF MELANOMA

Background: It is often difficult to differentiate early melanoma from benign pigmented le sions of similar clinical appearance. Objective: Our purpose was to develop a computer image analysis system that has t he potential for use as an adjunct to the clinical distinction of mela noma from less serious pigmented lesions. Methods: The system, consist ing of a hand-held device incorporating a color video camera and color frame grabber mounted in a microcomputer, was used in a pigmented les ion clinic. Analysis software extracted features relevant to the size, color, shape, and boundary of each lesion, and these features were co rrelated with clinical and histologic characteristics on which standar d diagnoses of skin tumors are based. For discriminant analysis based on image analysis measurements, equal prior probabilities were assigne d to two specified diagnostic groups, namely melanoma and ''other pigm ented lesions,'' most of which were melanocytic nevi. Results: In a 20 -month period, video images of 164 unselected pigmented lesions for wh ich complete diagnostic data were available were successfully captured using the camera. Sixteen of 18 melanomas, and 89% of pigmented lesio ns overall, were correctly classified by the image analysis system, co mpared with 83% based on clinical gradings of lesion characteristics. Conclusion: Computer image analysis has the potential to provide a val uable diagnostic aid that could enable clinicians to make highly sensi tive and specific diagnoses of early, curable melanoma.