CHICAGO- Almost comparable to expert radiologists at detection of breast cancer in women using screening mammograms is an artificial intelligence Google system; it also shows promise at reducing errors according to researchers in U.S. and Britain.
On Wednesday, the study was posted in the journal Nature and it shows how Artificial Intelligence (AI) can improve breast cancer screening accuracy. This is a relevant finding as breast cancer affects one of eight women globally.
According to the American Cancer Society, around 20% of breast cancers are missed by radiologists in mammograms. Over a 10-year period, half of all women who get screenings done are diagnosed with a false-positive result.
The study was developed with Alphabet Inc’s DeepMind AI unit which merged with Google Health in September and the finding represent a major advance in early breast cancer detection, according to Mozziyar Etemadi, one of the co-authors from Chicago’s Northwestern Medicine.
The team consisted of Imperial College London and Britain’s National Health Service’s researchers and they trained the system to screen through tens of thousands of mammograms to identify breast cancers.
The performance of the system was then compared to results from 3,097 mammograms from the United States and 25,856 mammograms from the UK.
The study finally showed that the identification of cancers by the AI system reported a similar degree of accuracy as expert radiologists and also reduced the incidence of false-positives by 5.7% in the U.S. group and by 1.2% in the British group.
The number of false-negatives, where the mammograms were wrongly diagnosed as normal, was also cut down by 9.4% in the U.S. group and by 2.7% in the British group.
The way that the mammograms were read reflects the resulting differences. In the U.S., the tests are done at a frequency of every one or two years and a single radiologist reads the results. In Britain on the other hand, tests are done once every three years and the results are read by two radiologists. If the two radiologists disagree, then a third is then consulted.
AI vs Expert Radiologists
In another test, the AI system was pitted against a team of six radiologists and it was seen that the system outperformed the doctors are detecting the breast cancers accurately.
These results are in line with findings from several researcher groups which used AI to detect cancers better in mammograms, including one from Connie Lehman’s work, who is the chief of Harvard’s Massachusetts General Hospital’s breast imaging department.
The idea of using computers to aid in cancer diagnostics is not new and computer-aided detection (CAD) systems are seen in mammography clinics commonly. However, these CAD programs haven’t made their way and improved clinical practice performance.
Lehman said that the issue with current CAD programs is that they are trained to see as human radiologists do. In contrast, AI computers can detect cancers with results from several mammograms.
AI, therefore, has the potential to exceed human capacity in the identification of subtle cues which the human brain and eye cannot perceive.
The tool can be used to guide doctors in making a well-informed decision.
It does have a few limitations- for starters, the tests were mostly done using the same imaging equipment and many of the U.S. group’s patients had confirmed breast cancers.
(Photos syndicated via Reuters)
This story has been edited by BH staff and is published from a syndicated field.