WEDNESDAY, April 17, 2024 -- For detection of malignant melanoma (MM), performance is more consistent when clinicians' work is supported by artificial intelligence (AI), according to a review published online April 8 in Cancers.

Ian Miller, from Southern Cross University in Bilinga, Australia, and colleagues conducted a systematic review to report the performance metrics of commercially available convolutional neural networks tasked with detecting MM. Data were included from 16 articles reporting MM, with 1,160 melanomas detected and 33,010 nonmelanoma lesions.

The researchers found that for classification of melanoma, performance of market-approved technology and clinician performance were highly heterogeneous, with sensitivity, specificity, and accuracy ranging from 16.4 to 100.0 percent, 40.0 to 98.3 percent, and 44.0 to 92.0 percent, respectively. When clinicians worked together with AI, less heterogeneity was seen, with sensitivity, specificity, and accuracy varying from 83.3 to 100.0 percent, 83.7 to 87.3 percent, and 86.4 to 86.9 percent, respectively.

"The implementation of AI in clinical practice to assist in the early detection of melanoma remains promising, particularly with the prospect of improved diagnostic accuracy," the authors write. "However, the heterogeneity of studies reporting on AI performance for classifying melanoma in clinical practice indicates it currently cannot be recommended as a reliable tool for clinicians."