Artificial intelligence could soon be used to diagnose melanoma more cheaply and accurately based on an artificial intelligence algorithm developed by WA researchers.
Findings published in the journal of the American Medical Association (AMA), led by an international team of researchers from the Harry Perkins Institute of Medical Research and Europe, found an AI algorithm was on par with trained dermatologists in identifying malignant pigmented skin lesions.
The algorithm was put to the test using smartphones and DSLR cameras with internet access to the algorithm and found to detect melanoma at similar levels to trained specialists.
The study also found that more than half of the malignant melanoma tissue detected by the A.I. were “in situ”, or stage 0, and were less than one millimetre deep, indicating the program’s potential for detecting early-stage lesions which might not be noticed.
Melanoma is currently the fourth most commonly diagnosed cancer for Australians, estimated to account for 11% of all cancer diagnoses in 2019. This places Australia as the country with the highest incidence rate of melanoma in the world.
When detected early, the relative five-year survival rate for melanoma patients is approximately 95%. However, as the five-year survival rate decreases to 8-25% for later diagnosis, there is pressure to identify malignant tissue in patients early.
Lead author of the findings, Perkins biostatistician Michael Phillips, said that due to this pressure, GPs and even dermatologists were often not confident enough in their examinations to rule out melanoma.
As a result, he said a high proportion of suspicious skin lesions removed from patients’ skin turned out to be benign.
“With improved accuracy our algorithm could reduce the unnecessary expense of pathology which costs enormous amounts of money, save patients an uncomfortable procedure and remove the need for specialist equipment and training on that equipment,” Mr Phillips said.
The study was funded by the creators of the AI program, medical device start-up Skin Analytics Limited.
