The latest medical innovation using artificial intelligence could help extend specialist healthcare beyond large hospitals and into small clinics and even homes. Researchers have developed a computer algorithm that can recognise skin cancer from photographs of skin lesions as efficiently as dermatologists can.
A group of dermatologists, microbiologists and computer engineers at Stanford University built a deep learning algorithm based on the Google Inception v3, which is a neural network algorithm. Such algorithms run computer programs in a manner similar to the way neurons in the human body process information. The research team ran 1,30,000 images of skin lesions from more than 2,000 diseases through the computer program. They then ran the algorithm head-to-head with 21 board-certified dermatologists to review hundreds of skin lesions to check whether they are malignant or benign. Neither the algorithm nor the doctors had seen the images previously.
In a study published in the journal Nature in January, the researchers showed that the algorithm identified skin cancer lesions about as well as the doctors did. Like the doctors, the program could also make distinctions between types of skin lesions telling the difference between keratinocyte carcinomas, which are the most common human skin cancer, and seborrheic keratosis that are benign growths.
This might be the most robust automated system developed yet to recognise skin lesions. While the program will need to undergo tougher test before being used for real diagnoses, the implications of an algorithm successfully identifying skin cancer are immense especially if it can be loaded and run on a smartphone, as one member of the research team tells Nature in this video.