Physics-based reaction-diffusion model predicts stroke infarct growth from routine MRI
maier_ak · x · 2026-09-11
Researchers from the University of Zurich and GIK Institute published a MICCAI Satellite Events paper, "An Imaging-Informed Reaction-Diffusion Model of Infarct Growth," exploring whether a classic physics-based PDE driven directly by routine MRI can predict infarct and penumbra evolution after stroke.
- Clinical context: thrombectomy decisions hinge on predicting how much brain tissue will die; current ML algorithms are accurate but opaque, while electrophysiological models need dozens of parameters not measurable from clinical MRI.
- The study tests whether the reaction-diffusion equation, driven by already-acquired imaging data, captures enough infarct dynamics to be useful.
- Author Andreas Maier adds caveats: Dice 0.46 is only an upper bound since reaction rate and threshold were fitted to each patient's 90-day outcome; the prospective model would lack those labels, the cohort is just 29 patients, no recanalization modeling, and no comparison to deep-learning segmenters.
Related event: Physics-Based Reaction-Diffusion Model Improves Stroke Lesion Prediction(2 posts)→
More from Research
- X-AuT: Progressive Audio-Encoder Compression for Speech LLMs via Cross-Scale Distillation — XPENG-AI · 2026-09-11
- Beihang's UniH3 Unifies Hierarchical Homogeneity and Heterogeneity for Medical Image Restoration — Beihang · 2026-09-11
- TempCloze Benchmark Tests Whether Video-LLMs Can Identify Missing Middle Video Segments — hkuhk · 2026-09-11
- Dev trains video diffusion model from first principles with motion-first approach — pixlpa · 2026-09-11
- Two AACL-IJCNLP papers probe how fine-tuning reshapes LLM internals and why WiC stays hard — Bollegala · 2026-09-11
- Stanford and Together AI paper: hybrid local-cloud routing cuts AI cost and energy by 60-80% — rohanpaul_ai · 2026-09-11