Skin cancer is one of Australia’s most widespread cancers, with over 80% of new cancer cases every year being attributed to skin cancer alone. Melanoma, the deadliest form of skin cancer, accounts for around 10% of new cancer cases diagnosed, which is around 15,000 new cases every year. With skin cancer happening in such high numbers, early detection and diagnosis is more important than ever.
This is why any technology that can help doctors offer more accurate and earlier diagnosis can make a real and deep impact on people’s lives in Australia. One of the most promising avenues of research is Artificial Intelligence (AI) which is already making fantastic strides in improving doctors’ approach to skin cancer detection.
In 2018 an international group of researchers from Germany, USA and France, developed and trained a convolutional neural network (CNN) to detect skin cancer by showing it more than 100,000 images of malignant melanomas and benign moles (also known as nevi). By training the CNN to differentiate between the two the researchers created an artificial intelligence that was very accurate in detecting potential skin cancer in patients.
Once the model was trained, the researchers invited 58 dermatologists from across the world to partake in a contest to see how accurate the CNN was compared to diagnosis by a doctor. In order to assess how experience affects the accuracy of diagnosis, the doctors were chosen from three groups; those with less than two years experience, those between two and five years experience and those with more than five years of experience.
300 images that had not been seen previously by the CNN or the doctors were used. The images of skin cancer were chosen to be difficult to diagnose, with lesions,, that to an untrained eye could have been a melanoma or just a regular mole.
The results after the test was run were surprising. The CNN correctly detected 95% of melanomas, while the doctors accurately detected 86.6%. The CNN also misdiagnosed less benign moles as melanomas in the trial, which means that it showed greater sensitivity and specificity.
On the doctors end their performance improved when they were given more clinical information and more experienced doctors performed better than less experienced doctors. However, the CNN still outperformed them in both cases, which highlights how artificial intelligence could provide a more standardised approach to skin cancer detection than detection by humans.
Despite these promising tests, researchers do not see the CNN completely replacing dermatologists, but rather, for it to act as another tool in the doctor’s repertoire. Another consideration is that while the CNN did outperform doctors, it was also done in a very specific setting, using high resolution images at 10 fold magnification, specifications which might be difficult to fulfil every time in a clinical setting, such as when the mole is in a difficult to image area like the scalp, fingers and toes.
So while for the moment AI is not going to take over the clinical diagnosis of melanomas, it will certainly become one of the major tools used by dermatologists to help them accurately diagnose skin cancer, allowing them to make better decisions on how to approach treatment for patients.