AI is better at detecting skin cancer than doctors

A new study shows that a form of artificial intelligence known as a deep learning convolutional neural network (CNN) is better than dermatologists at diagnosing skin cancer.
30 May 2018

More than 76,600 people in the U.S. are diagnosed with melanoma each year. Source: Shutterstock

The possibility of robots outperforming humans is something that seems to becoming increasingly possible. And according to new research, artificial intelligence (AI) may be better than trained humans at detecting skin cancer.

In a study carried out by an international team of researchers, experienced dermatologists were pitted against a machine learning system – known as a deep learning conventional neural network (or CNN) – in order to see which method was more effective at detecting malignant melanomas.

The results – published in the journal Annals of Oncology – found that most dermatologists were outperformed by the CNN.

The CNN was tested against 58 dermatologists from 17 countries. Over half were classed as “experts”, with over five years experience, 19 percent had two – five years’ experience, and 29 percent were beginners with less than two years experience in the field.

The study:

The dermatologists were shown 100 images of skin lesions and asked to make a diagnosis based on their judgment on whether it was a malignant melanoma or a benign mole.

58 dermatologists were involved in the study. Source: Shutterstock

Following this, they were also asked their opinion on how they would manage the condition (such as surgery, a short-term follow-up, or no action required).

After four weeks, the researchers gave the dermatologists clinical information about the patient, including age, sex, the position of the lesion, as well as close-up images. Again, they were asked to make a diagnosis and asks how they would manage the case.

A set of 300 images of skin lesions were also shown to the CNN. This tech is an artificial neural network inspired by the biological processes that happen when the brain’s neurons are connected to each other and respond to what the eye sees.

A CNN is capable of machine learning and teaches itself from what it has seen in order to get better.

The results:

Apon first viewing of the images, the dermatologists correctly detected an average of 87 percent of melanomas and successfully identified an average of 73 percent of non-malignant lesions.

On the other hand, the CNN correctly identified 95 percent of melanomas.

According to the report, dermatologists improved their diagnosis after they were given additional information about the patients along with the images.

Here, they were able to accurately diagnose 89 percent of malignant melanomas and 76 percent of benign moles.

However- they were still outperformed by the AI CNN system, which made decisions based solely on the images.

“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,” study author Professor Holger Haenssle, senior managing physician in the Department of Dermatology at the University of Heidelberg in Germany, said in a statement.

Could AI help diagnose increasing cases of skin cancer?

According to reports, in recent years, incidences of both non-melanoma and melanoma skin cancers have been on the rise.

Over 76,600 people in the US are diagnosed with melanoma each year and according to statistics from the U.S. Centers for Disease Control and Prevention, more than 9000 people die from it.

In cases of Melanoma early detection is vital for successful treatment. Unfortunately though, in many cases, diagnosis is made when the cancer is in advanced stages- making it more difficult to treat.

While AI should not replace doctors, it could be an incredibly useful tool in helping them to diagnose skin cancer more quickly and efficiently.

However, before AI can become a standard clinical tool, a number of issues would need to be addressed.

This includes the challenge of imaging melanomas in certain locations of the body, as well as the difficulty in training AI to detect less typical cases of melanomas.