A team of researchers from the Stanford’s Artificial Intelligence Laboratory employed a deep learning algorithm developed by Google (GoogleNet Inception v3 CNN architecture) to create an automated system for the classification of skin cancer. The algorithm that was already able to recognise objects from 1.28 million images, was trained using nearly 130,000 clinical images of more than 2,000 diseases. In a paper published on Nature, the researchers described how they employed over 370 biopsy-confirmed images to assess the artificial intelligence performance against 21 board-certified dermatologists. Notably, the algorithm matched the sensitivity and specificity of the experts’ opinion in classifying correctly keratinocyte carcinomas, melanomas classification and benign lesions. The new system paves the way of accessible and cost-effective applications that will enable recognition of skin lesions with the use of such devices as smartphones.
What is your opinion? Discuss on Among Doctors!
Image Credit: Stanford/Matt Young