Deep learning predicts receptor status on MRI of breast cancer brain metastases

A 3D deep learning model on post-contrast T1 MRI predicted breast cancer brain metastasis receptor status (micro-averaged AUROC 0.74; ER 0.57, PR 0.79, HER2 0.64). Proof-of-concept only, no clinical utility.

• 241 post-contrast T1-weighted MRIs from 142 patients; mask-free 3D ResNet ensemble with asymmetric focal loss and saliency maps.
• Held-out cohort contained only four HER2-negative patients, so threshold-dependent HER2 metrics are prevalence-driven; AUROC more appropriate.
• Authors frame as proof-of-concept methodological contribution, not performance gain; requires larger multi-center validation.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.