Why AI is changing mammography forever

AI SCIENCE

Why AI is changing mammography forever

A study in Nature Medicine explored how AI affects cancer detection and recall rates in mammography screening.

Breast cancer screening reduces deaths, but improving accuracy could further lower recall rates, reduce missed cancers, and improve treatment.

Currently, two radiologists interpret most mammograms, with extra reviews needed for difficult cases.

This process is repetitive and time-consuming, and with screenings expanding to more age groups, radiologists’ workloads are set to grow.

AI offers a way to ease these challenges. Research shows AI can be as accurate as, or better than, radiologists in spotting cancer.

Earlier studies, however, had small samples and lacked consistency in radiologists and equipment.

This study focused on AI’s role in a German breast cancer screening programme for women aged 50–69, analysing data from July 2021 to February 2023, with 461,818 mammograms reviewed by 119 radiologists.

In brief:

• The AI group had a breast cancer detection rate (BCDR) of 6.7 per 1,000, compared to 5.7 per 1,000 in the control group.

• AI marked 59.4% of mammograms as normal, reducing the workload significantly.

• Positive predictive values (PPVs) for recalls and biopsies were higher in the AI group, improving diagnostic confidence.

Fewer recalls, more results

The AI system, Vara MG, flagged normal cases and highlighted suspicious ones.

Its safety net aimed to catch missed cancers and reduce unnecessary recalls by prompting radiologists to double-check flagged findings.

Out of 260,739 mammograms read with AI and 201,079 without, the AI group had slightly lower recall rates and detected 17.6% more cancers.

Biopsies were 8.2% higher in the AI group, but these had better accuracy, with a PPV of 64.5% compared to 59.2% in the control group.

AI also detected more cases of ductal carcinoma in situ (DCIS), a non-invasive cancer.

However, this raises concerns about overdiagnosis, as DCIS doesn’t always lead to invasive cancer.

Researchers stressed the need to study rejected safety net cases further, as these could reveal missed cancers or show where unnecessary recalls were avoided.

In summary, using AI in mammography is safe and effective.

It reduces workloads while improving cancer detection, though more follow-up is needed to understand its long-term effects on overdiagnosis and missed cancers.

AI turning ‘What ifs’ into ‘We got this.’