Dermoscopy Explainable Intelligence Performance in Evaluating Melanoma and Non-Melanoma Skin Cancer

To assess the correlation between the Dermoscopy Explainable Intelligence (DEXI) score and histopathologically confirmed melanoma and non-melanoma skin cancer in a cohort of patients.

This retrospective review included 178 patients who underwent total body photography by the Vectra WB360 system (Canfield Scientific Inc), sequential digital dermoscopy imaging, and subsequent biopsy of 1 or more skin lesions between May 1, 2023, and August 30, 2024. By use of an artificial intelligence-generated scoring system trained on more than 66,000 annotated dermoscopy images known as DEXI, the dermoscopy images were assigned a score from 0.0 to 10.0 calculated to assist in malignancy assessment. The histopathologically confirmed diagnosis of biopsied lesions was then compared with its associated DEXI score. Average DEXI scores by diagnosis were determined, and a clinically appropriate cut point for DEXI score was calculated.

Of the 261 cases, 160 were benign with an average corresponding DEXI score of 2.5, 65 cases were non-melanoma skin cancer with an average DEXI score of 7.8, and 14 cases were melanoma with an average DEXI score of 6.0. DEXI scores from malignant lesions were statistically significantly higher than those from benign lesions. Using 3.0 as a cut point for DEXI score, we observed a sensitivity of 93.7% and a specificity of 75% in differentiating benign from malignant disease.

The DEXI score was overall effective at differentiating benign from malignant disease. Larger prospective multicenter validation will be required to further evaluate for clinical use.