Enabling DCIS subtyping: leveraging foundation models for robust grading and molecular biomarker scoring
AI analysis of routine breast cancer slides could spare low-risk patients from surgery by reliably identifying those safe for monitoring instead.
A foundation model deep learning pipeline trained on digitized H&E slides accurately predicts molecular biomarkers (ER, HER2) and grade in DCIS across two independent national cohorts, directly supporting the LORD trial's active surveillance eligibility criteria. With NPV of 0.86 in the Dutch cohort, this tool could reduce overtreatment of low-risk DCIS by reliably identifying patients safe for surveillance rather than surgery or radiation.
What the study was
- Study design
- Multicenter deep learning validation study
- Population
- Women with DCIS undergoing pathological evaluation; Dutch multicenter dataset + UK external validation
- Sample size
- 1146
- Category
- Diagnostics
- Maturity
- Validated
- Journal
- NPJ Breast Cancer
Why it surfaced
Multicenter validated AI pathology tool directly addresses DCIS overtreatment problem; UK external validation confirms generalizability; linked to active LORD trial for de-escalation decision support
A plain-language summary of published research — not medical advice. Talk to a clinician about your own care.