The Cancer Research UK has classified skin cancer as the fifth most common cancer in the world with thousands of new cases every year. Scientists say that the best way to prevent skin cancer is early detection. With that in mind, a team of researchers from Stanford University developed a new system that would detect skin cancer at its early stages using artificial intelligence.
The statistics from the Skin Cancer Foundation reveals that 5.4 million cases of nonmelanoma skin cancer are treated every year in the United States while one person dies every 52 minutes because of melanoma, the deadliest type of skin cancer.
Initially, skin cancer is detected by visual examination but with the new system developed by the Stanford researchers, that would soon become the task of artificial intelligence. The system is based on image recognition which could be developed for smartphones thereby allowing easy access to screening as well as providing low-cost methods for detecting skin cancer at its early stages.
The scientists used the deep learning algorithm created by Google to built the new system. According to Andre Esteva, an electrical engineering Ph.D. student at Stanford and the co-author of the study, they fed the system with 127,000 images of skin lesions that encompass all types of skin cancer. Each of these images is also individually labeled.
After that, they trained the algorithm to pick out and identify patterns and relationships. Once the training is done, the machine can identify new and unsorted data given to it.
In order to test the efficacy of their system, the scientists test it by presenting it with around 2,000 new images of different skin lesions. These images have been already biopsied by dermatologists. The system is on par with the experts in detecting different types of skin cancers from benign skin lesions to melanomas.
Esteva and his colleagues said that the system needs further testing. However, it already holds a lot of promise not only in skin cancer detection but in other fields of medicine as well.
The study is published in the journal Nature.
