Accuracy of Computer Diagnosis of Melanoma

Noninvasive diagnosis of melanoma with tensor decomposition-based feature extraction from clinical color image

“…In spite of best efforts of researchers, the accuracy of the noninvasive diagnosis of the melanocytic lesions is still far from ideal and differentiating a malignant melanoma from benign melanocytic lesions without histological examination represents a challenge [9]. One of the most widely used methods for preliminary diagnosis based on visual inspection is the so called ABCDE rule.…”

“…The possibility to identify melanoma from visual information motivated development of methods based on image analysis. In this regard, in [9] a comparative performance analysis of twelve studies is presented, with accuracy exceeding the one attained by visual inspection. Use of a computer-aided diagnostic system with high level of sensitivity and specificity (typically above 90%) could provide a second-opinion to the dermatologist and also reduce the number of unnecessary biopsies.…”

“…Use of a computer-aided diagnostic system with high level of sensitivity and specificity (typically above 90%) could provide a second-opinion to the dermatologist and also reduce the number of unnecessary biopsies. Systems for automatic discrimination of skin lesions based on dermoscopic imaging are described in [9], [13], [14], [15], [16].…”

“…While automatic systems based on dermoscopic and high spectral resolution MSI/HSI systems show great potential, it was demonstrated in [9], as well as recently in [17] and [18], that automatic diagnosis of melanoma is possible using color RGB autofluorescent images. Presented diagnostic methods utilizing RGB images achieved sensitivity in range 80-94% and specificity in range 46-95%.…”

“…Presented diagnostic methods utilizing RGB images achieved sensitivity in range 80-94% and specificity in range 46-95%. Results of the twelve studies reported in [9] as well as results from [17] and [18] are summarized in Table 1.…”