In a study, a new AI system was found to be more effective than doctors in detecting the presence of breast cancer in mammograms.
The earlier breast cancer is detected, the higher the chances of survival. This is why it is recommended that women have regular mammograms. This tool is very effective, but it is not perfect. It is estimated today that one in five breast cancers is not detected by this method. Over a ten-year period, it is also estimated that one in two women will have a false positive diagnosis at some point.
For several years, several teams of researchers have been developing artificial intelligence in order to improve the accuracy of these results . Developed by Google Health, a new system has proven to be particularly effective in analyzing mammograms.
Trained initially on 91,000 breast x-rays, the system has recently been tested on two large datasets involving another 28,000 mammograms. The first was for American women and the second for English women.
In the US dataset, the software diagnosed 5.7% fewer false positives compared to a human doctor. The system also recorded 9.4% fewer false negatives , suggesting that it detected more breast cancers missed by the physician.
Concerning the English data, the mammograms this time were examined by two separate radiologists. The margin for human error has therefore been reduced. Nevertheless, the AI still performed better with 1.2% fewer false positives and 2.7% fewer false negatives.
A second part of the study consisted of pitting six American radiologists against the AI system against the analysis of 500 randomly taken mammograms. Again, the system detected several cancers missed by the six experts. On the other hand, a case detected by the latter had been missed by artificial intelligence. So the tool is not perfect, but it still shows amazing results.
Ultimately, for researchers, the idea is not to replace doctors, but to assist them in order to accelerate the process of current diagnoses .
“This is a great demonstration of how these technologies could support human experts “, explains Dominic King, one of the co-authors of the study. “The AI system could flag a possible abnormality on an X-ray and ask the doctor to check behind “.
In general, the development of AI is rather well perceived by professionals in the medical sector. According to an article recently published by the MIT Technology review, 80% of the 900 respondents believe that these systems have already reduced the burnout of healthcare workers. Artificial intelligence would also free up time for doctors to spend more time with their patients.
Source
Related articles:
