Researchers from the Karolinska University Hospital in Stockholm tested three computer-aided diagnostic systems using artificial intelligence algorithms. The research material consisted of mammography data from a large population of women undergoing regular screening. The goal was to test whether the technology can detect subtle changes that are not yet visible to the human eye.

The analyzes showed that the AI ​​algorithm scores were significantly higher in patients whose cancer was later confirmed. Women who remained healthy scored low on the systems. This suggests that the technology recognizes patterns predicting the development of the disease long before its clinical manifestation.

Predicting diagnosis six years from now

“Approximately 20% of breast cancer cases show signs on mammograms that AI can detect about six years before diagnosis. Our study confirms the potential of AI to detect cancer signs on mammograms in some cases much earlier than radiologists,” says Dr. Fredrik Strand from Karolinska University Hospital, author of the publication in Radiology.

Algorithms have previously shown promise in estimating five-year risk of disease, the researchers note. They were also effective in detecting patients at risk of developing cancer between screening tests. This time, the team decided to delve further and test how far into the past AI can go in detecting changes.

Thousands of mammograms under the microscope of algorithms

Swedish specialists analyzed the systems’ ability to recognize changes visible in images taken ten years before diagnosis. The analysis included almost 90,000 samples. Mammograms were performed on more than 31,000 patients from four Swedish regions. The entire material covered a ten-year period of observation.

Each image was additionally annotated by two radiologists, allowing comparison of human and machine performance. During the study period, doctors diagnosed cancer in 38.5% of the participants. The results of the algorithm were then compared with these final clinical diagnoses.

The authors emphasize that an early warning signal allows the implementation of measures adapted to a specific patient. This could mean, among other things, an increase in the frequency of follow-up examinations. This saves time, which in oncology can be crucial for treatment success.

“The goal of this study is to enrich the growing literature on the use of artificial intelligence in breast cancer screening and to demonstrate how it can play a role in the earlier detection of this cancer,” concludes Dr. Strand. “Analysis of AI scores in screened individuals over time could provide insight into how early detectable changes occur, potentially allowing for earlier intervention,” he adds.

This article comes from the Ringier Media publishing partner website. The content and data contained in it were taken without editorial intervention.