11 Sept. 2023. Results of a clinical trial show one radiologist using software with artificial intelligence can detect more breast cancer cases in mammograms than current reviews by two radiologists. Findings from the trial conducted by researchers at Karolinska Institutet in Stockholm, Sweden using a cancer diagnostics system by Lunit Inc. in Seoul, South Korea, appear in Friday’s issue of the journal The Lancet Digital Health.
Mammography remains a basic tool for breast cancer screening, which requires interpretation of the breast images by trained radiologists. In Europe and other regions, current medical practice calls for two radiologists to review the images. However, the authors cite their own earlier study showing a wide variation in interpretations by 110 radiologists of mammogram images compared to a subsequent diagnosis of breast cancer. In addition, say the authors, trained radiologists for interpreting mammograms are in short supply. Thus the researchers evaluated a system using A.I. algorithms to reduce the need for two radiologists to interpret mammogram images.
The study called ScreenTrustCAD used Lunit’s Insight MMG system that the company says employs deep learning algorithms constantly updated with new results to inspect mammogram images for suspicious lesions indicating cancer. Lunit says the system improves cancer detection even in dense or fatty breast tissue that can make mammograms more difficult to analyze. The company says its Insight systems have received FDA clearance in the U.S. and CE marks in Europe. In addition, the authors earlier assessed three A.I. commercial diagnostics technologies and rated the Insight MMG system as the top performer for use in this study.
Fewer false positives and reduced workloads
The researchers evaluated human and A.I. image interpretation across a wide population in Sweden to find if algorithm-assisted detection is at least equivalent to a diagnosis made by radiologists. The study team tested mammograms taken at a hospital in Stockholm from 58,344 women age 40 to 74 in 2021-22, with images from 55,581 included for analysis. All mammogram images were first reviewed by radiologists, then assigned for review either by another radiologist or the Lunit system. The researchers looked primarily for diagnosis of breast cancer within three months of the mammogram, with detection rates for a radiologist plus A.I. system, compared to review by two radiologists. The team also compared assessments by a single A.I. system and evaluations made by two radiologists plus the A.I. system.
Results show assessments by two radiologists detected breast cancer in 250 women, while one radiologist and the A.I. system found 261 cancer cases or four percent more than the conventional two-radiologist method. Using the A.I. system alone detected cancer in 246 women, within the statistical confidence interval of two radiologists, while adding A.I. to the two radiologists detects 269 cancer cases. The findings also show fewer false positive readings by the combination of radiologist and A.I. system compared to two radiologist, and reduced workloads on radiologists.
The authors indicate the human and machine assessments complement each other in detecting cancer in mammograms. “A.I. and humans perceive images slightly differently,” says radiologist and first author Karin Dembrower in a Karolinska Institutet statement, “which creates a synergy that improves our chances of detecting cancer.” Senior author Magnus Strand, a cancer pathologist at Karolinska Institutet adds, “It’s clear to us that for screening mammography, one A.I.-supported radiologist is a better alternative than two radiologists without A.I.”
“A.I. is redefining cancer screening standards,” notes Brandon Suh, CEO of Lunit in a company statement, adding “This study signifies a milestone in health care, ushering in an era where A.I. seamlessly complements and elevates the standards of breast cancer screening.”
More from Science & Enterprise:
- NIH Challenge Seeks Early Alzheimer’s Indicators with A.I.
- Biotech, A.I. Companies Model Animal Studies/Clinical Trials
- A.I. Analytics Detect Heart Failure in Single Ultrasound View
- Digital Biomarkers Shown to Highlight Parkinson’s Disease
- Neuro Biomarkers, Home EEG Assessed in Upcoming Trial
We designed Science & Enterprise for busy readers including investors, researchers, entrepreneurs, and students. Except for a narrow cookies and privacy strip for first-time visitors, we have no pop-ups blocking the entire page, nor distracting animated GIF graphics. If you want to subscribe for daily email alerts, you can do that here, or find the link in the upper left-hand corner of the desktop page. The site is free, with no paywall. But, of course, donations are gratefully accepted.
[wpedon id=”42724″ align=”center”]
* * *
You must be logged in to post a comment.