A study by researchers from Sweden's Karolinska Institute has revealed that a single radiologist working with artificial intelligence can detect more cases of breast cancer than two radiologists working together to examine mammograms.
The results of the study, which indicated that artificial intelligence is now ready for use in breast cancer detection, were published in The Lancet Digital Health in early September and written by EurekAlert.
In the past 30 years, mammography has played an important role in reducing deaths from breast cancer. However, challenges to tackling cancer, including the lack of radiologists and the inability to diagnose all cancer cases, persist.
A large number of studies have shown that AI can help address these challenges.
Karen Dembroer, the first author of the study, who works in the Department of Oncology and Pathology at the Karolinska Institutet, noted that artificial intelligence and humans look at images a little differently, which together has a synergistic effect, increasing the likelihood of cancer diagnosis.
Read all scan
Radiologists usually read each examination, and this method is standard in care. Women who may need more screening are then called in to diagnose breast cancer.
In this study, the results of the two radiologists were compared with that of different numbers of radiologists powered by artificial intelligence.
The study was conducted at the Kapio St. Göran Hospital in Stockholm, Sweden, from April 2021 to June 2022. The study included more than 55,40 women screened between the ages of 74 and <>.
The traditional method of reading X-rays was able to capture 250 cases. The number of cases detected was unexpectedly larger when two AI-assisted radiologists read the images, with 269 cases detected.
A single AI-powered radiologist detected 261 cases in the same group, while AI alone detected 246 cases, making it statistically no less than two radiologists combined.
Lead researcher Frederick Strand, a radiologist in the Department of Oncology and Pathology at the Karolinska Institutet, noted that reading images by a single radiologist powered by artificial intelligence increased the number of detected cases by 4 percent and also halved the time a doctor needs to read images.
The results of the study also showed that AI assistance to the radiologist or the use of artificial intelligence alone reduced the likelihood of misdiagnosis for women without cancer, a mistake in which women without cancer are called for further examinations, which causes them suffering and increases their costs without need.
Dr. Strand says it has become clear that reading radiology images through a single AI-powered radiologist has become a better option than reading images with two radiologists.
In this study, it was statistically verified that artificial intelligence improves the diagnosis of breast cancer, which did not happen in a previous study conducted at Lund University.
He adds that the use of artificial intelligence heavily in initial examinations will not obviate the need for a radiologist to judge before asking the patient to return for other tests and take a biopsy if cancer cells are suspected.
AI is ready
Dr. Strand adds that this study indicates that artificial intelligence is ready for controlled use in reading radiology images, but it is necessary to choose an artificial intelligence program that has been well examined on a number of images taken using the same device, which will then read its images and ensure that it continues to be monitored after clinical use. In the long term, artificial intelligence is expected to be able to evaluate most mammography images.
Starting in June this year, mammography at Cabo Saint Goran Hospital was evaluated by an AI-powered doctor, freeing up time for doctors to dedicate to cancer patients.