AI system almost unerringly diagnoses skin cancer 31.01.2017
AI system almost unerringly diagnoses skin cancer

Doctors at Stanford University in liaison with programmers have developed a system with artificial intelligence, which allows distinguishing ordinary moles from skin cancer. The system diagnoses melanoma better than experienced doctors do.

The number of people that suffer from skin cancer increases every year. According to statistics, last year 5.5 million people died of malignant neoplasms of this type in the US. And only 14% of patients with advanced stage of melanoma were saved. Therefore, a system that helps to unmistakably detect a need for an urgent surgery or to confirm that it is just a new pigment spot is really priceless.

The novelty is very simple to use. The patient should take a picture of a suspicious mole on smartphone, then based on deep convolutional neural network the system analyzes the image and compares it to big number of mole and melanoma photos uploaded to the database.

Tests of this smart program confirmed its superior diagnostic capabilities. The system’s precision in recognition of melanoma was 72%, while doctors confirmed the diagnosis only in 66% of cases. Thus, the installation of the program on a smartphone can replace a visit to a dermatologist. However, doctors are skeptical of this novelty and claim that the system will not be able to distinguish rare malignancies such as non-pigmented melanomas.

AI-industry news and cases for you