58 dermatologists from 17 countries, against an AI. The winner? The AI.
Called the Convolution Neural Network, it was shown over 100,000 pictures of skin cancers and began its deep learning program to process and spit out its takes on what the picture could mean.
Of all the forms of skin cancer, Melanoma is the most deadly of the bunch, expecting a death count of about 10,000 in the US. But when diagnosed they are usually in the late stages making it hard to treat and to recover. But maybe the turnout could drop if the AI were to be inputted in the diagnoses process.
The AI has a set up that can be related to a child’s. It looks at a picture and can identify it through machine learning, and as it views more and more pictures it will improve its accuracy of identifying the cancer.
To further test the effectiveness of the AI against normal dermatologist deductions, each were given 100 pictures of skin lesions, and of course they were told to diagnose the pictures and offer follow-up actions.
The AI got 95 percent of the diagnoses correct while the dermatologists only got up to 86 percent. Although in the second stage of the testing, dermatologists went up to 89 percent, but the AI still performed better.
Some believe however, that the dermatologists underperformed because they didn’t take the study seriously because they knew for a fact that they weren’t diagnosing actual people. This means that there are more discrepancies in the study that was performed and that further research must be performed.