A recent study published in JAMA Network Open investigated whether a commercial artificial intelligence (AI) tool—originally designed for breast cancer detection—could analyze screening mammograms to predict the risk of developing breast cancer years before clinical diagnosis. Researchers studied nearly 350,000 mammograms from 116,495 women, using AI-based cancer detection scores as indicators of future cancer risk rather than immediate diagnosis.
Key Findings: AI Detects Subtle Risk Patterns Early
The study found that the AI algorithm assigned higher cancer detection scores to breasts that eventually developed cancer, even 4 to 6 years prior to diagnosis. These elevated scores contrasted significantly with those from breasts that remained cancer-free. This suggests that commercial AI tools can identify women at higher risk well before symptoms or clinical signs appear, potentially enabling more personalized and timely screening interventions.
Breast-Level Tracking Reveals Risk Trends
As per the News Medical and Lifesciences, detailed breast-level analysis showed progressively widening differences in AI scores between the two breasts of women who later developed cancer. Specifically, score differences increased from 21.3 points two screening rounds before diagnosis to 79.0 points at the time of detection, highlighting AI’s sensitivity to subtle changes over time.
Background: Challenges in Traditional Mammography Screening
Breast cancer remains the most commonly diagnosed cancer among women, accounting for approximately 25% of female cancer cases worldwide, with over 2.3 million new diagnoses and 670,000 deaths annually (2022 data). Although routine mammography screening significantly improves survival by detecting cancers early, it depends heavily on radiologist expertise and often misses malignancies that develop between screening intervals, leading to delayed diagnosis and worse outcomes.
Advancements: AI Enhances Risk Assessment Accuracy
Recent developments in AI have allowed the generation of detailed neoplasm scores by analyzing minute variations in mammograms. These scores assist clinicians in identifying breast cancer risk earlier and more accurately than traditional methods. Previous research speculated that AI could predict future breast cancer based on subclinical mammographic features, but this large-scale cohort study offers robust scientific validation of that hypothesis.
About the Study: Methodology and Data Sources
The study used a retrospective cohort design following STROBE guidelines. It sourced data from Norway’s BreastScreen program, which conducts biennial digital mammography on women aged 50–69. Each participant had at least three mammograms, independently assessed both by radiologists and an AI tool called INSIGHT MMG (version 1.1.7.2), developed in collaboration with Lunit Inc. The AI provided continuous cancer detection scores from 0 to 100, with higher values indicating greater risk. The researchers analyzed these scores and their bilateral differences across screening rounds and correlated them with subsequent breast cancer diagnoses.
Study Results: Predictive Power of AI Scores
The AI scores showed strong predictive value years before cancer diagnosis. For example, breasts destined to develop cancer had mean scores twice as high as healthy breasts (19.2 vs. 7.1) even four years before diagnosis. Women who developed breast cancer exhibited consistently higher mean AI scores and larger differences between their breasts over time, compared to those who remained cancer-free. Sensitivity analyses confirmed these findings, reinforcing the reliability of AI in risk prediction.
Moreover, the AI tool performed as well as or better than established clinical risk calculators, such as Tyrer-Cuzick and BCRAT models, in discriminating short-term breast cancer risk.
Important Clarifications and Limitations
It is crucial to note that the AI algorithm did not “diagnose” cancer years in advance; rather, it indicated increased risk based on subtle image features. The study does not prove that early AI-based risk detection improves patient survival or reduces costs—those outcomes require further investigation.
Additionally, the study’s retrospective design, focus on a single commercial AI model, and predominantly White Norwegian cohort limit generalizability. The authors call for prospective studies and validation in more diverse populations to optimize clinical use.
Conclusion: Toward Personalized Risk-Based Screening
This large population-based study demonstrates that commercial AI tools can detect subtle mammographic features associated with future breast cancer risk years before diagnosis. Such AI-driven risk assessment holds promise for personalized screening strategies, allowing earlier intervention for high-risk women. However, further research must confirm these findings across diverse groups and establish best practices for integrating AI risk scores into routine clinical care.