A study conducted in Sweden shows that AI can diagnose breast cancer 20% more accurately in mammography examinations than the usual analysis performed by doctors, writes Politico.
The study examines the use of AI in breast cancer screening for the first time. Preliminary results published in the medical journal Lancet Oncology on Tuesday afternoon, the 1st of August, found that AI-assisted analysis of mammography results involving one or two radiologists was as successful as two radiologists, and detected up to 20% more cancers. case.
The time spent on reviewing examinations has also decreased significantly – up to 44%.
The trial in Sweden is still ongoing, and more than 80 thousand women have participated in it so far. For half of the participants, the results were reviewed by only two radiologists, while for the other half, the mammography results were analyzed by a radiologist and an AI. The exception was cases when the AI showed a particularly high risk – then a second diagnostic specialist was also involved.
AI capabilities are increasingly being used in medicine, but at the same time, there are concerns about how algorithms are “trained” and tested in this environment.
The European Union is planning stricter regulation of AI use, and the European Drug Agency is evaluating the risks and benefits of using the technology in drug development.
However, the lead author of the article, Kristina Lång from Lund University in Sweden, pointed out that, despite the positive results,
preliminary information alone is not yet sufficient to confirm the use of AI in interpreting mammography results.
Lång and colleagues await the further results of the trial, which will determine whether this technology is worth using more widely.
Lång said the biggest benefit of AI right now might be less workload for radiologists. AI assistance could mean one radiologist is needed instead of two, allowing more patients to be treated.
Stephen Duffy, professor of cancer screening at Queen Mary University of London, pointed out that reducing the workload of radiologists is a critical issue. He added, however, that
the possibility that the AI could overestimate even harmless masses is a cause for concern.
The Swedish study is also the first to investigate the use of AI in analyzing mammography results. A paper published in European Radiology in January notes that trials are also essential to monitor the safety of AI systems, which can also make unpredictable and undetectable errors.
Read also: Tony Blair: AI will change everything