In what can be considered a knee jerk to qualified radiologists, a study has shown that Artificial Intelligence (AI) can detect breast cancer from screening mammograms with the same level of accuracy as professional radiologists. Researchers also found that the AI could reduce false-negative errors.
The study is a result of the employment of Alphabet Inc's (Google's parent company), DeepMind AI. It included researchers from National Health Service, Britain, Google Health, and Imperial College London.
According to the American Cancer Society, radiologists failed to spot nearly 20% percent of breast cancer cases from mammograms. Also, among women who receive mammograms over a period of ten years, nearly half of them show false-positive results.
Training AI to identify breast cancer
The AI system was trained using thousands of mammograms. The goal was to teach the system to identify the mammograms that showed signs of breast cancer. Following this, the researchers took the confirmed results from mammograms— 3,097 from the United States and 25,856 from the United Kingdom—and compared it with the AI's score.
Results showed that the AI performed at the level of qualified radiologists when it came to accuracy. It also showed a marked decrease in the number of false-positive results—by 1.2 percent in the UK group and 5.7 percent in the US group. The AI also reduced the number of false negatives—wrong classification of tests as positive—by 4 percent in the US-based group and by 2.7 percent in the UK-based group.
"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," Ara Darzi, co-author of the study, told the BBC.
Making AI and radiologists lock horns
The researchers matched six radiologists against the AI in a different test. They found that the AI was able to beat them when it came to the accurate detection of breast cancers. Connie Lehman, chief of the breast imaging department at Harvard's Massachusetts General Hospital, who was not a part of the study but has worked in the discussed area, told Reuters that the results of the study are in alignment with the findings of several studies that focussed on the use of AI to improve breast cancer studies.
She said that current computer-aided detection (CAD) systems are trained to identify signs that radiologists notice. However, by using AI, computers can learn to detect cancers using the actual results of mammograms numbering in thousands.
The work has the prospects to "exceed human capacity to identify subtle cues that the human eye and brain aren't able to perceive," added Lehman.
A long way to go
The US-based group consisted of many patients who were already diagnosed with breast cancer. Also, most of the tests were conducted using similar kinds of imaging implements. Since it is still a study and has not seen practical implementation in clinical conditions, it is premature to look at the full-fledged implementation. The study is yet to go under the scanner of regulatory approval.
Dr Lisa Watanabe, chief medical officer of CureMetrix, told Reuters that, "AI software is only helpful if it actually moves the dial for the radiologist." Watanabe, who was not related to the study also added that the researchers are yet to demonstrate that the tool can improve patient care.
While agreeing that further studies are required, Mozziyar Etemadi, a co-author of the paper, told Reuters, that this study can serve as a step forward in the potential for the early detection of breast cancer.