How accurate is AI at reading mammograms compared with radiologists, and what did large screening studies like MASAI in

Recent large-scale studies suggest that AI-supported mammography screening is highly effective, performing at least as well as, and in some metrics better than, human radiologists.

The MASAI (Mammography Screening with Artificial Intelligence) trial, published in The Lancet Oncology in 2023, is the most prominent study on this topic. Here is a summary of what that study and the broader field have found:

Findings from the MASAI Trial

The MASAI trial was a large, randomized controlled trial conducted in Sweden involving over 80,000 women. It compared two methods: standard double-reading (where two radiologists read each mammogram) and AI-supported screening (where AI pre-screened the images, and radiologists reviewed them).

• Detection Rate: The AI-supported screening detected 20% more cancers than the standard double-reading method.
• Workload: The use of AI significantly reduced the screen-reading workload for radiologists by 44%, as the AI could effectively filter out clear cases.
• Recall Rates: Despite the increase in cancer detection, the study found that the number of false-positive recalls (women called back for follow-up who did not have cancer) did not increase significantly.
• Safety: The study concluded that AI-supported screening was safe and effective, with no evidence that it missed cancers that would have been caught by human-only review.

How AI Compares to Radiologists

Current evidence suggests that AI acts as a powerful "second pair of eyes" rather than a replacement for human experts. The general consensus is:

• Sensitivity: AI is excellent at flagging subtle anomalies (like early-stage microcalcifications or architectural distortions) that might be overlooked by a fatigued human reader.
• Consistency: Unlike humans, AI does not experience fatigue, distractions, or variations in mood, providing a consistent baseline of performance regardless of the time of day.
• The "Human-in-the-Loop" Factor: Radiologists remain essential for interpreting the context of a mammogram. They can synthesize the AI's findings with a patient’s medical history, clinical symptoms, and prior imaging, which AI cannot yet do comprehensively.
• Complementary Strengths: Most experts believe the best performance occurs when a radiologist uses AI as a decision-support tool. The AI handles the initial triage and feature detection, allowing the radiologist to focus their expertise on complex cases and final clinical decisions.

In short, the MASAI study provided strong evidence that integrating AI into screening workflows can increase cancer detection rates while reducing the burden on clinical staff, without compromising patient safety or increasing unnecessary follow-ups.