An early-stage study discovered that artificial intelligence (AI) outperformed doctors with extensive training and experience in detecting breast cancers, reducing doctors’ mammogram reading workload by almost half. However, this doesn’t imply that computers will soon determine cancer diagnoses in hospitals. The study, published in The Lancet Oncology on Tuesday, demonstrates the safety and potential effectiveness of AI in breast cancer detection, potentially surpassing the capabilities of human doctors.
Previous research has shown AI’s usefulness in predicting breast cancer risk, but those studies primarily used models or retrospective data. This new research is believed to be the first randomized control trial comparing AI-assisted breast cancer detection with the efforts of well-trained human radiologists alone.
The study involved more than 80,000 women in Sweden who underwent mammograms between April 2021 and July 2022. Half of the women had their mammograms read by AI before being analyzed by a radiologist, while the other half had their mammograms read by two radiologists without AI assistance. Both groups of radiologists were highly experienced.
The group whose scans were reviewed by AI and a radiologist had 20% more cancer cases detected compared to the group reviewed by two radiologists without AI assistance. Overall, AI-supported screenings had a cancer detection rate of 6 per 1,000 women screened, compared to 5 per 1,000 with the standard approach. Importantly, the AI did not increase false positives, where a mammogram is diagnosed as abnormal despite no cancer being present.
Additionally, the AI-assisted group enjoyed a 44% reduction in reading workload. While the exact time saved was not measured, researchers estimated that if radiologists read around 50 mammograms per hour, using AI could potentially save a single radiologist four to six months in reading about 40,000 screening exams compared to two radiologists working alone.
Dr. Kristina Lång, an associate professor of radiology diagnostics from Lund University in Sweden and co-author of the study, emphasized that the most significant benefit of AI is its potential to alleviate the excessive reading burden on radiologists.
In Europe, guidelines recommend two radiologists to screen a mammogram, while the US lacks a specific standard, leading to varying workload issues in different countries. Nevertheless, both Europe and the US face a scarcity of radiologists, which AI-supported breast cancer detection might help address, while enhancing radiologists’ performance.
As the global population ages and requires more imaging, the demand for radiologists is expected to rise. Many radiologists view AI’s possibilities as welcome advancements rather than threats to their job security. However, further research is required to confirm the technology’s effectiveness before widespread implementation.