Summary: A study from Washington University School of Medicine highlights an AI-based mammogram analysis method that detects breast cancer risk 2.3 times more accurately than traditional methods, offering equitable and improved early detection across diverse populations.

Key Takeaways

  1. Enhanced Accuracy: The AI-based method predicts breast cancer risk 2.3 times more accurately than traditional questionnaire-based approaches by analyzing up to three years of mammogram data.
  2. Equitable Performance: The algorithm effectively detects risk across diverse racial groups, addressing gaps in representation and ensuring accuracy for underrepresented populations, including Black women.
  3. Broader Application: Researchers aim to make the technology widely available by pursuing patents and testing its effectiveness across additional racial and ethnic groups to support personalized care.

————————————————————————————————————————————————————————

A study from Washington University School of Medicine in St. Louis presents an artificial intelligence-based method for analyzing mammograms, offering a major leap in predicting breast cancer risk. By analyzing up to three years of previous mammograms, this approach identifies high-risk individuals 2.3 times more accurately than traditional questionnaire-based methods that rely on factors like age, race, and family history.

Published in JCO Clinical Cancer Informatics, the study leverages machine learning to detect subtle changes in breast density, texture, calcification, and asymmetry that even trained radiologists might miss. These changes provide valuable insights into early cancer development.

“Our goal is to improve early detection and prevention,” says senior author Graham A. Colditz, MD, DrPH, associate director of prevention and control at Siteman Cancer Center in St. Louis. “Better risk prediction can help guide preventive measures for high-risk women.”

AI Algorithm Enhances Accuracy Across Groups

The researchers trained the AI algorithm using over 10,000 mammograms from Siteman Cancer Center patients (2008–2012), followed through 2020, during which 478 women developed breast cancer. They validated the model with mammograms from more than 18,000 women screened at Emory University (2013–2020), where 332 cases were diagnosed.

The results showed that women in the high-risk group were 21 times more likely to develop breast cancer within five years than those in the low-risk group. The new method correctly identified 53 cases per 1,000 high-risk women, compared to 23 cases using older methods.

Importantly, the algorithm was designed to be effective across diverse racial groups, including robust representation of Black women, who are often underrepresented in such models. Its accuracy held up across all groups, paving the way for equitable screening advancements.

Toward Broader Adoption

The researchers are pursuing patents and licensing to make this technology widely available. They also aim to ensure the algorithm works well for women from varied racial and ethnic backgrounds, including Asian, Southeast Asian, and Native American populations.

“This study highlights the potential of AI to revolutionize breast cancer screening by identifying risks invisible to current methods,” says Debbie L. Bennett, MD, chief of breast imaging at Mallinckrodt Institute of Radiology. “It’s a promising step toward better, more personalized care.”