Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Predicting Breast Cancer Treatment Response with Integrated Imaging and Pathology AI
A new explainable AI system combining MRI and tissue pathology predicts breast cancer response to neoadjuvant therapy before treatment begins, enabling personalized treatment strategies.
Background
Response to neoadjuvant therapy varies significantly among breast cancer patients. Identifying responders before treatment could enable personalized planning and reduce unnecessary toxicity. This study developed a graph neural network integrating radiomic features from MRI with pathomic features from biopsies to model intratumoral heterogeneity across spatial scales.
Key Findings
- Integrated radio-pathomic model achieved superior performance (AUC 0.95 training, 0.91 external test) compared to imaging-only (AUC 0.89/0.84) and tissue-only (AUC 0.87/0.83) approaches
- Tissue features dominated treatment response predictions while imaging features predicted non-response, showing complementary modality strengths
- Top-ranked nodes exhibited elevated tumor-infiltrating lymphocyte densities, a recognized histopathologic marker of treatment response
- Model performance remained robust across molecular subtypes and tumor sizes (AUC 0.89–0.93)
Why It Matters
Pretreatment prediction enables early identification of treatment non-responders, allowing clinicians to tailor therapy intensity and surgical planning accordingly while reducing overtreatment toxicity. The system’s explainability at topology, node, feature, and cross-modal levels supports informed clinical decision-making.
Limitations
This was a retrospective dual-center study. The approach requires both pretreatment MRI and whole-slide biopsy images for model implementation.
Original paper: Radiopathomic Graph Deep Learning for Multiscale Spatial-Contextual Modeling of Intratumoral Heterogeneity to Predict Breast Cancer Response to Neoadjuvant Therapy. — Radiology. Artificial intelligence. 10.1148/ryai.250760
