Researchers from Google, Northwestern University, the NHS, and Imperial College London have discovered that Google’s Deepmind AI outperforms radiologists in detecting breast cancer. The researchers trained the AI on thousands of mammograms, which resulted in the tech being able to identify breast cancer cases better than the radiologists who had made the initial assessment. In the fight against cancer, AI is winning – sometimes.
The Study
When analysing 18,000 patients from the US, the research showed that there was a 5.7% reduction in false positives, and a 9.4% reduction in false negatives. In the UK, the data showed a 1.2% reduction in false positives, and false negatives reduced by 2.7%. The researchers had a much bigger database to work from in the UK, screening more than 100,000 patients.
The researchers then pitted the AI system against a group of six highly skilled radiologists. They were tasked with interpreting 500 cases, all randomly selected. Here the AI outperformed the radiologists too, however it did miss some cases that the team had highlighted as positive for cancer. So, in the fight against cancer, AI is winning – sometimes.
As the study’s abstract notes, the system is “……capable of surpassing human experts in breast cancer prediction.” However, it’s unlikely that AI will replace human radiologists in the field anytime soon.
Machine learning algorithms are more than capable at completing repetitive tasks (like going over thousands of x-rays), it’s not quite ready to start popping up in hospitals and clinics.
According to Dr Hugh Harvey, managing director of digital health consulting firm, Haridan Health, “AI is perfect to support breast cancer screening. Machines are perfect for doing mundane, high-volume tasks.” He continued, “All of this early research is all well and good. To actually get it deployed into the clinic is a whole lot of work.”
An AI system will also still need to be approved by the FDA, and would likely have to connect with a hospital’s health records, and would need constant surveillance (human or otherwise) to ensure that its findings are maintaining their accuracy.
Will it be implemented?
Frankly, the study does have its limitations. Yes, researchers had a large and diverse data set, but all of the images were from a single manufacturer’s mammography system. The actual demographics of the study (as in the patients) was also not too diverse, and the study should have used a better variety of the population to enable any real accuracy. This technology will also have to be tested in clinical trials before it’s ever implemented.
“The real world is more complicated and potentially more diverse than the type of controlled research environment reported in this study,” wrote Dr. Etta Pisano, chief research officer for the American College of Radiology, in an analysis of the research. She added that earlier versions of this technology, like computer-aided cancer detection, showed good initial results. But when it came to real-world application, the systems just weren’t up to scratch.
Still, what benefits can this tech have? Can it reduce patient mortality, or make radiologists more efficient? Will it reduce costs? So far, it looks promising, but we haven’t seen anything concrete.
“Those are the papers we’ll see in the next couple of years,” Harvey noted. “The only industry we can look at to give us an estimate of how long (approval) takes is the pharmaceutical industry. People predict the digital sector will go faster, but we still have to do these prospective studies.”
It won’t be in the field anytime soon
While this report is very promising, the actual technology won’t be in the field anytime soon. Yes, in the fight against cancer, AI is winning – sometimes. But, really, only sometimes. Will this technology have enough real world benefits to pass clinical trials, or will it never see the daylight. AI has the power to change lives, and save lives, but we’ll never see this project come to fruition unless they expand on their research. Still, this research has certainly highlighted one important thing – there’s nowhere AI can’t reach.
