RÉSUMÉ
OBJECTIVE: To demonstrate the safety and effectiveness of MelaFind, a noninvasive and objective computer-vision system designed to aid in detection of early pigmented cutaneous melanoma. DESIGN: A prospective, multicenter, blinded study. The diagnostic performance of MelaFind and of study clinicians was evaluated using the histologic reference standard. Standard images and patient information for a subset of 50 randomly selected lesions (25 melanomas) were used in a reader study of 39 independent dermatologists to estimate clinicians' biopsy sensitivity to melanoma. SETTING: Three academic and 4 community practices in the United States with expertise in management of pigmented skin lesions. PATIENTS: A total of 1383 patients with 1831 lesions enrolled from January 2007 to July 2008; 1632 lesions (including 127 melanomas-45% in situ-with median Breslow thickness of invasive lesions, 0.36 mm) were eligible and evaluable for the study end points. MAIN OUTCOME MEASURES: Sensitivity of MelaFind; specificities and biopsy ratios for MelaFind and the study investigators; and biopsy sensitivities of independent dermatologists in the reader study. RESULTS: The measured sensitivity of MelaFind was 98.4% (125 of 127 melanomas) with a 95% lower confidence bound at 95.6% and a biopsy ratio of 10.8:1; the average biopsy sensitivity of dermatologists was 78% in the reader study. Including borderline lesions (high-grade dysplastic nevi, atypical melanocytic proliferations, or hyperplasias), MelaFind's sensitivity was 98.3% (172 of 175), with a biopsy ratio of 7.6:1. On lesions biopsied mostly to rule out melanoma, MelaFind's average specificity (9.9%) was superior to that of clinicians (3.7%) (P=.02). CONCLUSION: MelaFind is a safe and effective tool to assist in the evaluation of pigmented skin lesions. TRIAL REGISTRATION: clinicaltrials.gov Identifier: NCT00434057.