Microsoft's new AI model for breast MRI screening slashes false positive results by 25% while maintaining high accuracy for detecting actual cancers.

Rahul Somvanshi

Unlike traditional AI, the Fully Convolutional Data Description (FCDD) model learns what normal breast tissue looks like and flags anything unusual as potentially concerning.

Photo Source: Cancer Research UK (CC BY-SA 4.0)

The technology creates visual heatmaps showing suspicious areas with 92% pixel-wise accuracy, building trust with radiologists who make the final diagnosis.

Photo Source: PUBS

Among 9,738 breast MRI exams analyzed, the model performed exceptionally well in low-prevalence scenarios where only 1.85% of scans contained cancer—similar to real-world screening rates.

Photo Source: Microsofy Blog

For the nearly 50% of women with dense breast tissue, this AI tool offers particular value by improving detection in tissue that traditionally makes abnormalities harder to spot.

Photo Source: Anna Tarazevich (Pexels)

"We are very optimistic about this new AI model," says Professor Savannah Partridge of University of Washington, noting it works with both full and abbreviated MRI protocols.

Photo Source: National Cancer Institute

The open-source code is available on GitHub as researchers prepare for wider clinical trials, emphasizing that AI will support rather than replace radiologists in cancer detection.

Photo Source: UN Women Asia (CC BY-NC-ND 2.0)

Regulatory approval remains the next crucial step, with the model likely to pursue a De Novo pathway rather than the simpler 510(k) clearance that existing breast MRI AI technologies use.

Photo Source: National Cancer Institute (CC0)

Future trials aim to include more diverse populations, as the development cohort was over 80% White with varying breast tissue density distributions.

Photo Source: Microsoft Blog

Next Story

Tumor Medication Linked to Gambling and Hypersexuality - Karmactive