DeepMind’s New AI Outperforms Doctors At Spotting Breast Cancer

Navin Bondade 0 Comments Artificial Intelligence, Data Science, Deep Learning, Machine Learning

Breast Cancer is the global leading cause of cancer-related deaths in women, alone in the US the average risk of a woman developing breast cancer in her life is about 12%, i.e 1 in 8 women.

Google’s DeepMind has demonstrated a study that can in the future bring down these numbers. The team has developed an AI system that can detect the presence of breast cancer with accuracy more than a doctor.

For the development and testing of the system, Google’s DeepMind made a collaboration with the team of UK researchers from Imperial College London at Google Health.

The team of researchers has built a system that can successfully identify breast cancer in X-ray mammograms & the system also capable of providing an independent, automated and immediate second opinion.

Mammography is the most common type of screening for breast cancer. It uses low energy X-rays to detect typical masses associated with breast cancer.

Screening is an important first step in identifying breast cancer the earlier it is spotted, the more and better types of treatment are available to the patient.

From the last two years, the researcher from Cancer Research UK Imperial College, Northwestern University, Royal Surrey County Hospital, and Google Health are putting efforts to develop a deep-learning system on two different datasets of breast scans, one from the US and one from the UK.

The study involves, using the AI programs trained on 2D and 3D mammography images from nearly 91,000 of the breast cancer screenings of women from the UK and the US.

The scientists trained an ingenious set of three different neural networks that each looked at mammograms at different levels of detail.

The particulars of this set-up of deep learning are fascinating and perhaps represents the state of the art in combining machine learning networks.

The system was first trained on de-identified mammograms from 76,000 British women, using Cancer Research UK’s OPTIMAM dataset, as well as 15,000 scans from the US.

Once trained, the algorithm was tested on 25,000 scans in the UK, and a further 3,000 in the US.

Four images from each mammogram were pulled into a neural network, spitting out scores for three different models between zero and one, with the latter a high risk of cancer.

The team of researchers said that the DeepMind’s system was able to identify those that had their cancer confirmed within the following year via a tissue biopsy oeamr subsequent X-rays.

They then compared the AI system’s performance with the actual results from a set of 25,856 mammograms in the United Kingdom and 3,097 from the United States.

The study showed the AI system could identify cancers with a similar level of accuracy to expert radiologists.

At the same time, it reduced the number of false-positive results by 5.7 percent in the American patients and 1.2 percent in the British patients.

It also cut the number of false negatives, where tests are wrongly listed as normal, by 9.4 percent in the American group, and 2.7 percent in the British group.

Additionally, Google’s researchers simulated its performance in automating the double-reading screening process employed in U.K. hospitals.

“This is another step along the way of trying to answer some of the questions that will be critical for us to actually deploying this in the real world,” says Dominic King, director and UK lead of Google Health.

“This is another step closer to trying to deploy this type of technology safely and effectively.”

By only bringing in a second human when the AI and the first clinician disagreed, they found the program was able to reduce the workload of the backup reader by 88%, while still maintaining the standard of care and providing on-the-spot feedback.

The study claims the DeepMind algorithm performs better than a single radiologist and is “non-inferior” versus two.”The model performs better than an individual radiologist in both the UK and the US,” Kelly says.

“The AI system natively produces a continuous score that represents the likelihood of cancer being present.”

“To support comparisons with predictions of human readers, we thresholded this score to produce analogous binary screening decisions,” where “threshold” in this case means picking out a single point to compare.”

“For each clinical benchmark, we used validation set to choose a distinct operating point; this amounts to a score threshold that separates positive & negative decisions” Dominic King, director of Google Health said.

The study has some limitations. Most of the tests were done using the same type of imaging equipment, and the U.S. group had a lot of patients with confirmed breast cancers.

Importantly, the team has not yet shown that the tool improves patient care, said Dr. Lisa Watanabe. She is chief medical officer of CureMetrix, a company whose AI mammogram program won U.S. approval last year.

She noted that AI is only helpful if it creates noticeable progress for radiologists.

Professor Ara Darzi, one of the authors of the paper and the director of the Cancer Research UK Imperial Centre, said he had not expected to see such an impressive result from the AI system.

“This is one of those transformational discoveries you have in your hand, which could disrupt the way we deliver screening in terms of improving accuracy and productivity,”

“These results highlight the significant role that AI could play in the future of cancer care,” Cancer Research UK chief executive Michelle Mitchell told. “Embracing technology like this may help improve the way we diagnose cancer in the years to come.”

“This is another step along the way of trying to answer some of the questions that will be critical for us to actually deploying this in the real world,”