Pitfalls of an automated dermoscopic analysis system in the differential diagnosis of melanocytic lesions.

Dermoscopy plays an important role in the diagnosis of pigmented lesions, particularly in the differential diagnosis of early-stage melanoma. Dermoscopy systems that aim to enable automatic "unmanned-without physician" diagnosis are becoming increasingly common. We aimed to investigate the reliability and weaknesses of diagnosis programs. Furthermore, we attempted to determine whether such programs are superior to diagnosis by a physician, compared to histopathological assessment. The images stored in the DermoGenius ultra-computerized dermoscopy system of the Dermoscopy Unit between January 2008 and December 2008 were surveyed retrospectively. Dermoscopic images made prior to excision of 77 lesions from 51 patients verified by histopathology were reviewed. Nineteen patients were men and 32 were women. Mean age was 35.5 years. Diagnosis by a clinician or automatic analysis revealed that 23 (30%) of the lesions were atypical (dysplastic) nevi, 22 (29%) were compound nevi, 10 (13%) were dermal nevi, 8 (10%) were malignant melanomas, 7 (9%) were common nevi, 6 (7%) were junctional nevi, and 1 (1%) was a blue nevus. Compared to histopathological diagnosis, considered the gold standard, the sensitivity of the automated analysis program was 96.6%, its specificity 14.9%, and its diagnostic accuracy 47%. For the clinician, the values were 100% for sensitivity, 66.7% for specificity, and 95% for diagnostic accuracy. Based on histopathological results, the diagnostic accuracy of the physician was higher than that of the automatic analysis program. Therefore, errors are inevitable when an inexperienced physician assesses patients according to automatic program results.