Skin Cancer Diagnosis Using AI
Let us illustrate some unique qualities and benefits of AI with an example.
Could a machine distinguish skin cancer from a benign skin condition—acne, a rash, or a mole—by simply scanning a photograph? Can a machine do it as accurately as an expert dermatologist, or even better? A research group at the Stanford University set out to find out the answers.
Stanford researchers started with an existing deep learning algorithm for AI built by Google for image classification. Just like any dermatologist, an AI system has to be trained to become accurate at diagnosing skin cancer. In order to do this the Stanford researchers used a “teaching-set” of 13,000 validated image samples spanning 2,032 different diseases, where each image had been diagnosed and thus categorized by dermatologists as benign lesions, malignant lesions, or non-cancerous growths. The deep neural network of the AI system then scanned these images pixel by pixel, looking for the characteristics common to each diagnosis. The AI system learned to diagnose these images and compared its diagnosis with the “correct” answer provided by the dermatologists. Subsequently, then, based on how accurate it was, the AI system tuned itself with every sample to improve its accuracy for the next diagnosis. After iteratively learning by comparing its diagnosis with that diagnosed by dermatologists through the vast number of images, its diagnosis became as accurate as the best dermatologists. In a side-by-side comparison, using 2000 gold standard test set of images, the AI system outperformed expert dermatologists.
The training and tests were done with high-quality images. With this said, more work needs to be done to make the system work with images taken with a smartphone and via the Internet for the diagnosis. Skin cancer diagnosis done by AI system with smartphone images could have an immense impact on the cost savings and value for patients and cost savings. Each year, some 5.4 million new cases of skin cancer are diagnosed in the United States. The usual process for identifying the many varieties of the disease involve a visual examination of moles or other marks on the skin by a dermatologist. In cases of doubt, a biopsy is needed. Early detection greatly increases the chances of survival – a five-year survival rate for melanoma detected early on is around 97 percent; but when detected in its later stages, that figure falls to around 14 percent.
This example illustrates the power of using AI for medical diagnosis. The use of images and patterns is very common in monitoring and diagnosis of many diseases. A well trained deep learning AI is excellent at extracting information from images, videos and patterns to accurately make the right decisions. A number of companies are very active in using AI for diagnosis. IBM Watson, an IBM supercomputer that combines artificial intelligence (AI) and sophisticated analytical software for optimal performance as a “question answering” machine, has made major progress with AI in diagnosing cancer and health disorders. We believe that AI will play a significant role in medical diagnosis for a wide range of diseases – lowering cost, reducing the time to detect, and saving lives.