Artificial Intelligence outperforms doctors at detecting breast cancer in new study

A recent study published in Nature - one of the world’s top academic journals – suggests that artificial intelligence (AI) is more accurate than doctors in diagnosing breast cancer from mammograms.

The international team involved in the study included researchers from both Google Health and Imperial College London. They designed and trained a computer model on X-ray images from almost 29,000 women and the algorithm outperformed a total of six radiologists when interpreting mammograms individually.

The study found that AI performed on the same level as two doctors working together would.

Experts say that AI could improve detection as unlike humans, AI is tireless.

The current process for analysing such X-rays in the NHS involves two radiologists and in rare cases where they disagree, a third doctor is called upon to also asses the images.

Women aged between 50 and 70 are invited for NHS breast cancer screening every three years - those who are older can ask to be screened.

Study

Within this research study, the AI model was given anonymised images meaning the women could not be identified. The human experts, however, had access to the patient’s medical history, as they would in practice.

The study found that the AI model was as good as the current system of two radiologists analysing the images. However, when the AI model was compared to just a single doctor, it was actually found to be better at detecting cancer.

For example, comparing the AI model’s performance against one radiologist, there was a reduction of 2.7% in false negatives, where the cancer is missed. 1.2% in false positives i.e. when a mammogram is diagnosed as abnormal incorrectly.

Dominic King from Google Health said: "Our team is really proud of these research findings, which suggest that we are on our way to developing a tool that can help clinicians spot breast cancer with greater accuracy."

Cancer Research UK's OPTIMAM dataset provided the mammograms which were collected from St George's Hospital London, the Jarvis Breast Centre in Guildford and Addenbrooke's Hospital, Cambridge.

According to the BBC, it takes over ten years of training as a doctor and specialist to become a radiologist, capable of interpreting mammograms.

Shortage of radiologists in UK

Reading X-rays is critical but time-consuming work, and there is an estimated shortage of over a thousand radiologists across the country.

It is important to note that this was a research study and that as of yet, artificial intelligence has been confined to laboratories and not used in any clinics. When AI is eventually used in clinics, at least one radiologist would remain in charge of the diagnosis.

However, experts say that using AI would at least eliminate the need for two radiologists to analyse mammograms, therefor easing pressure on their workloads.

Director of the Cancer Research UK (CRUK) Imperial Centre and co-author of the report, Prof Ara Darzi, told the BBC: "This went far beyond my expectations. It will have a significant impact on improving the quality of reporting, and also free up radiologists to do even more important things."

The use of AI in clinics could speed up diagnosis, as images can be analysed within seconds by the computer algorithm.

Director of cancer intelligence and early diagnosis at CRUK, Sara Hiom, told the BBC: "This is promising early research which suggests that in future it may be possible to make screening more accurate and efficient, which means less waiting and worrying for patients, and better outcomes."

Helen Edwards, from Surrey, was diagnosed with breast cancer before she was eligible for screening at the age of 44.

She was a patient representative on the CRUK panel and had to decide whether to grant Google Health permission to use the anonymised breast cancer data.

Helen told the BBC: "Initially I was a bit concerned about what Google might do with the data, but it is stripped of all identifiers.

"In the long term this can only benefit women.

"Artificial intelligence machines don't get tired... they can work 24/7 whereas a human being can't do that, so to combine the two is a great idea."