Can AI Provide an Early Alert for Breast Cancer Before Diagnosis

Three commercially available AI radiology systems can identify early signs of breast cancer up to six years before a formal diagnosis, according to a study by Hickman et al published in Radiology. The research, led by Dr. Fredrik Strand of Karolinska University Hospital, analyzed data from over 88,000 mammograms to track cancer prediction scores over a decade.

How did the AI detect cancer signs early?

Researchers conducted a retrospective study using three AI-based computer-assisted detection (AI-CAD) systems. They tested these systems on mammogram data from the Validation of Artificial Intelligence for Breast Imaging (VAI-B) database, which includes volunteers from four regions in Sweden.

The study analyzed 88,963 mammograms from 31,394 patients over a 10-year period between January 2008 and April 2019. During this time, radiologist readers diagnosed 12,072 participants, or 38.5% of the group, with cancer.

The AI-CAD systems assigned prediction scores to the images. Scores were elevated on average for individuals who later received a cancer diagnosis and remained low for those who stayed cancer-free.

Did You Know? The Swedish national breast screening program invites women between the ages of 40 and 74 to participate in screening examinations every two years, with each mammogram traditionally read by two radiologists.

What were the specific results of the AI screening?

The AI systems achieved 90% specificity, meaning they effectively distinguished between true positive and true negative results. According to the study, the AI identified signs of cancer in 19.7% of individuals up to six years before their recorded diagnosis.

How breast cancer is detected | Fredrik Strand

Detection rates increased as the diagnosis date approached. The AI flagged signs in up to 25.2% of individuals four years before diagnosis and 39.3% of individuals two years before diagnosis.

Dr. Fredrik Strand noted that approximately 20% of breast cancer cases show mammographic signs visible to AI around six years before a diagnosis is made. He stated the study confirms AI’s potential to find signs earlier than radiologists.

Expert Insight: Samantha Carter suggests that the ability to identify high-risk scores years in advance could shift the clinical focus toward personalized surveillance. By flagging subtle changes that precede a clinical diagnosis, providers may be able to implement closer vigilance for specific patients.

How could this change future breast cancer screenings?

AI-CAD scores may help radiologists identify early mammographic signs of potential future cancers. Dr. Strand stated that analyzing these scores over time could provide insight into how early detectable changes arise.

A personalized approach to interpreting these individual scores could help clinicians identify patients who might benefit from closer vigilance. This shift in monitoring could potentially allow for earlier intervention.

The research adds to existing literature on AI’s ability to predict five-year breast cancer risks and identify women at risk for “interval” cancers that occur between regular screenings.

Frequently Asked Questions

How many years before a diagnosis can AI flag breast cancer signs?
According to the study, AI systems showed the potential to flag early signs up to six years before a diagnosis.

What was the specificity rate of the AI-CAD systems?
The systems achieved 90% specificity in distinguishing between true positive and true negative results.

Who participated in the study?
The study included 31,394 patients and 88,963 mammograms from the VAI-B database in Sweden.

How do you feel about the integration of AI into routine medical screenings?