Blog · ppj.es
AI in mammography: does not prolong reading and detects more cancers
Published on 19 August 2026
Title: AI in mammography: does not prolong reading time and detects more cancers — prospective alternating months study
Type: A (AI medical imaging / breast)
Tags: AI mammography, detection, screening, radiology, MARTE
PMID: 42593499
DOI: 10.1007/s00330-026-12793-0
Journal: European Radiology (2026)
Category: Oncology
⚖️ Transparency notice: this article was written with AI assistance and reviewed by the author, a medical oncologist.
The design that matters
Prospective “alternating months” study (August 2023–July 2024) at Yongin Severance (Korea): 4 radiologists, 4577 mammograms (screening + diagnostic), same commercial AI system integrated into workflow, with results shown or hidden depending on the month. This eliminates selection bias and isolates the effect of AI in real practice [PMID 42593499].
The numbers
| Metric | With AI | Without AI | p |
|---|---|---|---|
| Overall CDR (per 1000) | 22.3 | 11.5 | 0.005 |
| Screening CDR | — | — | no diff |
| Diagnostic CDR | 18.7% | 12.1% | 0.026 |
| Screening AIR | 9.5% | 8.4% | 0.293 |
| Diagnostic AIR | ↑ | — | higher with AI |
| Mean reading time | 65.0 s | 64.3 s | 0.723 |
AI did not prolong reading time (65.0 vs 64.3 s, p=0.723) and increased overall detection rate (22.3 vs 11.5/1000, p=0.005) [PMID 42593499].
Critical reading
- Only 4 radiologists, single center. Generalization to other settings (breast density, prevalence) is limited.
- Effect is greater in diagnostic than in screening. In screening there was no difference in CDR or AIR; in diagnostic AI increased detection BUT also abnormal interpretations (more potential false positives). AI helps more where there is more uncertainty.
- Identical reading time. Disproves the fear that AI slows down workflow — a practical argument for adoption.
- Alternating months design is strong (controls case-mix better than cross-sectional).
💡 Implications for consultation / MARTE
Evidence that AI in mammography is complementary, not substitutive: maintains radiologist pace and improves detection without time cost. The nuance: in diagnostic it also increases abnormal interpretations, so the AI threshold must be calibrated to avoid inflating biopsies. Connects with MARTE (our AI project on imaging reports): the value lies in the second reader, not in replacing the first.
Reference: Lee SE, et al. Eur Radiol. 2026. doi:10.1007/s00330-026-12793-0. PMID: 42593499.
— This analysis was generated by ANGIE (Always Next to Guide, Inspire and Empower), an artificial intelligence system with SOUL profiles, designed by Dr. Javier Pumares Pérez.
Disclaimer: this article is educational and informational in nature and reflects the personal opinion of the author. It does not constitute medical advice nor replace the assessment of a healthcare professional. If you have a health concern, consult your physician.