Medical-AI trial: predicting high/low risk and subtype of basal cell carcinoma from dermoscopy images with a CNN — compared with dermatologists (NCT07677124)

1Basal cell carcinoma and its subtypes

Basal cell carcinoma (BCC) is the most common type of skin cancer; most grow slowly, but if neglected or incompletely removed they can spread deeply in one area. Planning treatment hinges on distinguishing a higher-grade subtype (high-risk) from a lower one (low-risk). Dermatologists use dermoscopy, a tool that magnifies the skin to see patterns below the surface, but the final subtype is confirmed by pathology of a biopsy.

2Anticipating subtype from clinical and dermoscopic images

This study examines whether that distinction can be supported from images. Per the registry summary, it uses clinical photographs and dermoscopy images with a CNN (a deep-learning method strong at image recognition) to evaluate, as the primary objective, the performance of classifying BCC as low- or high-risk.

Secondarily, it predicts the histopathological subtype and compares the model with the assessments of dermatology physicians. Cases are histopathologically confirmed BCC from a dermatology archive, with pathology as the reference standard. It is retrospective, with a target of 2,500, and it is recruiting.

3Forming a view before the biopsy

Deciding treatment for basal cell carcinoma turns on whether the histological subtype is high risk. The study asks whether that judgement can be supported at the image stage.

Using clinical photographs and dermoscopic images, a convolutional neural network model is evaluated primarily on classifying BCC as low or high risk. Secondary aims cover predicting the histological subtype and comparing against dermatologists' assessments. Comparing against dermatologists matters in positioning AI as support for judgement in the field rather than as a replacement.