The world of artificial intelligence (AI) is expanding rapidly with breakthroughs happening each and every day. Computers are getting faster and more intelligent, the manufacturing process is becoming more efficient, and improvements are being seen all across the healthcare industry. 

One such success to emerge just recently is thanks to a group of researchers from Germany, France, and the USA. They’ve developed a form of AI that’s more accurate than human dermatologists when it comes to detecting skin cancer.


More than 100,000 images of malignant melanomas were shown to the deep learning convolutional neural network (CNN). Upon comparing the results of the AI’s diagnosis with that of 58 international dermatologists, the researchers found that the CNN missed fewer melanomas and had fewer misdiagnoses than the humans.

Explaining a little more about how the CNN is able to produce such accurate results is the first author of the study, Professor Holger Haenssle, senior managing physician at the Department of Dermatology, University of Heidelberg: “The CNN works like the brain of a child. To train it, we showed the CNN more than 100,000 images of malignant and benign skin cancers and moles and indicated the diagnosis for each image. With each training stage, the CNN improved its ability to differentiate between benign and malignant lesions.”

Of the 58 dermatologists taking part, 17 had less than two years’ experience, 11 had between two and five years, and 30 had more than five years experience under their belt. To begin with, each dermatologist was asked to diagnose a malignant melanoma or benign mole from 100 of the same dermoscopic images shown to the CNN. They were also to decide as to the right course of treatment (if any). Then, around one month later, they were given personal information (i.e. age, sex, and position of the lesion) about the people, shown close up images of the same 100 cases and then asked to give a re-diagnosis. 

Results from the study revealed that when it came to the first set of diagnoses’ the dermatologists had a success rate for detecting melanomas of 86.6% and 71.3% for non-malignant lesions. The CNN on the other hand, managed to successfully identify 95% of melanomas while identifying the exact same number (71.3%) of benign moles. The dermatologists did improve slightly in the second set of diagnosis with an accuracy rate of 88.9% for detecting malignant melanomas and 75.7% that were benign.


“The CNN missed fewer melanomas, meaning it had a higher sensitivity than the dermatologists, and it misdiagnosed fewer benign moles as malignant melanoma, which means it had a higher specificity; this would result in less unnecessary surgery,” said Haenssle. “These findings show that deep learning convolutional neural networks are capable of outperforming dermatologists including extensively trained experts, in the task of detecting melanomas.”

The number of people being diagnosed with malignant melanomas is on the rise with around 232,000 new cases being reported every year. If detected early enough, it can be cured. But often people aren’t diagnosed until it’s progressed and then it is much harder to treat. While Haenssle himself has been involved in various projects involving the early detection of melanomas over the past 20 years and made some significant advancements in the area, there’s always room for improvement.

Should dermatologists be worried that they will no longer be needed in their field of medicine? No. The CNN is not there to replace humans in diagnosing skin cancers. It’s simply there to act as an aid. “This CNN may serve physicians involved in skin cancer screening as an aid in their decision whether to biopsy a lesion or not. Most dermatologists already use digital dermoscopy systems to image and store lesions for documentation and follow-up. The CNN can then easily and rapidly evaluate the stored image for an ‘expert opinion’ on the probability of melanoma.”


Of course, the study was not without limitations. These include the fact that the dermatologists were in a fake setting where real decisions weren’t having to be made; the tests themselves didn’t cover the full spectrum of skin lesions; there were fewer images from non-Caucasians skin types, and not all doctors will follow the recommendation of a CNN they’ve yet to trust. It’s quite obvious that there’s still a lot to be done before AI will be accepted into mainstream clinical settings. But, it is no doubt, something that is coming in the near future.

More News to Read

Comments

comments