Machine versus man in skin cancer diagnosis
Menée à partir de 1 511 images dermoscopiques et auprès de 511 dermatologues, étudiants en dermatologie ou généralistes, cette étude internationale évalue, par rapport à la précision de l'être humain, celle d'algorithmes d'apprentissage automatique pour diagnostiquer divers types de lésions cutanées bénignes ou malignes
Skin examinations for diagnosis of skin cancer are facilitated by the use of dermatoscopy, a non-invasive skin imaging technique that amplifies features of pigmented skin lesions that are not easily discernable when examined by the naked eye. Since its introduction in the late 1980s, dermatoscopy has substantially increased diagnostic accuracy in the identification of melanoma. However, interpretation of dermatoscopic criteria remains a subjective procedure that requires specific and dedicated training. Several strategies based on artificial intelligence and machine learning, including deep learning networks, have been developed as aids for less experienced dermatologists or other physicians to classify dermatoscopic images into benign and malignant skin tumours, as well as more specific diagnostic categories, which in the long-term could revolutionise dermatology practice.
The Lancet Oncology , commentaire, 2018