Can AI Predict Breast Cancer Years Before Diagnosis?

AI breast cancer detection is rapidly transforming the landscape of oncology and preventive healthcare. Recently, researchers evaluated how artificial intelligence models flag high-risk scores on screening mammograms years before a clinical diagnosis. This study from BreastScreen Norway demonstrates that advanced algorithms identify cancer risk markers years before radiologists see visible signs.

The Efficacy of AI Breast Cancer Detection Models

Furthermore, the study analyzed retrospective data from over 130,000 screening examinations. Researchers evaluated two distinct AI models, including a commercial model and an in-house model. Consequently, both models assigned high-risk scores to women years before they received a formal diagnosis. Specifically, the commercial model selected 43 cases, while the in-house model identified 47. Interestingly, both models selected 29 of the same cases, showing significant overlap.

In addition, radiologists conducted a retrospective, informed review of three consecutive mammograms for each patient. These scans occurred four years prior, two years prior, and at the time of diagnosis. Therefore, this detailed review offered a unique timeline of cancer progression. The analysis revealed that AI markings correctly matched the future tumor location in up to 61% of cases four years prior. Conversely, human radiologists classified 89% of these early examinations as completely normal or showing only minimal non-specific signs.

Understanding the Evolution of Mammographic Features

Additionally, the research tracked how mammographic features changed over time. At the time of diagnosis, spiculated masses and densities with calcifications represented the most common findings. However, these typical signs were entirely absent two and four years earlier. This means the AI models identified subtle, non-visual patterns rather than traditional visible signs. Consequently, these findings highlight the limitations of human visual inspection alone.

Ultimately, these results suggest that AI can detect microscopic or sub-visual breast changes. For this reason, utilizing AI as a supportive tool could significantly improve patient outcomes. However, implementing this technology in clinical practice requires careful consideration. Since radiologists still classify most of these early scans as negative, a major challenge lies in managing false-alarm fatigue. Therefore, clinicians must establish optimal risk thresholds before widely adopting these models.

Frequently Asked Questions

Q1: How early can AI models identify breast cancer risk markers?

According to the study, AI models can identify breast cancer risk markers up to two and four years prior to a clinical screen-detected diagnosis.

Q2: Why do radiologists often classify these early mammograms as negative?

Radiologists often classify these scans as negative because the standard mammographic features, such as spiculated masses, are physically absent at that early stage. Therefore, the AI is detecting sub-visual, subtle changes that are invisible to the human eye.

Q3: Did the AI markings match the actual location of the future cancer?

Yes. Indeed, the AI markings correctly corresponded to the future cancer location in at least one mammographic view for up to 61% of the cases four years before diagnosis.

References

1. Martiniussen MA et al. Location of AI risk markers and associated mammographic features in screening mammograms obtained years before screen-detected breast cancer. Eur Radiol. 2026 Jun 18. doi: 10.1007/s00330-026-12689-z. PMID: 42313161.
2. Strand F et al. AI spots breast cancer signs years before diagnosis. Radiology. 2026 June.
3. Sermo. AI Breast Cancer Detection: What Physicians Need To Know. July 2025.