Artificial Intelligence Outperforms Dermatologists in Detecting Skin Cancer
A neural network was both more sensitive and more specific than a group of dermatologists in accurately diagnosing skin cancer.
Researchers have demonstrated that a form of artificial intelligence known as a deep learning convolutional neural network can diagnose skin cancer more accurately than professional dermatologists. The study appeared in the journal Annals of Oncology.
For the study, an international team of researchers trained a convolutional neural network (CNN) to identify skin cancer by feeding it more than 100,000 images of malignant melanomas and benign moles, along with the diagnosis for each image. Then they asked the CNN to diagnose a new set of images and compared its performance with that of a group of 58 dermatologists.
The CNN missed fewer melanomas and misdiagnosed fewer benign moles as malignant than the dermatologists.
The researchers say that improved diagnostic accuracy could lead to earlier diagnoses, fewer unnecessary procedures and overall better outcomes for patients. They see technologies such as CNNs not as a replacement for dermatologists but rather a tool to help physicians screen for skin cancer and aid in their decision whether or not to biopsy a lesion.
However, there are a number of hurdles before artificial intelligence comes to a clinic near you. These include the difficulty of providing clear enough images for potential skin cancer lesions at awkward locations such as the toes and scalp, and how to train a CNN to recognize atypical melanomas.