Can algorithms be trained to detect skin cancer with accuracy comparable to that of human oncologists?
If you are curious about technology, you must have heard of computer vision. Deep neural networks are taking image analysis to a whole new level. This includes medical imaging. Can algorithms be trained to detect skin cancer with accuracy comparable to that of human oncologists?
The answer is yes!
Neural networks are already capable of such diagnosis, with increasing accuracy. What makes it important is that early detection of skin cancer means high probability of full recovery.
In February 2017, Nature published a study by Stanford researchers. They developed an algorithm able to detect carcinoma, one of the most common malignancies, and melanoma, the most serious form of skin cancer (5% of images analyzed). The model proved to be as accurate as top medical experts.
Another interesting study was published in January 2018. A group of researchers at Yale trained an artificial neural network to estimate a person’s risk of developing melanoma. This predictive model can make a positive impact, as it significantly improves early detection of dangerous conditions.
What comes next?
Mobile applications for individual users. It’s already happening — go ahead and google AI-powered apps, such as Doctor Hazel. Millions of mobile device users around the world can provide enough training data, and in several years an AI personal oncologist on your smartphone may become a reality.
Worth the effort, wouldn’t you say?
