Brain Tumor Growth Inversion via Differentiable Neural Surrogates

Jonas Weidner, Lucas Zimmer, Ivan Ezhov, Michal Balcerak, Bjoern Menze, Daniel Rueckert, Benedikt Wiestler

Keywords: Neural Surrogates, Gradient-Based Optimization, Brain Tumor Models

TL;DR: We drastically accelerate the fitting of personalized biophysical brain tumor models to improve radiotherapy treatment planning.

Abstract: Personalizing biophysical brain tumor models to individual patients is computationally expensive due to the need for numerous iterative evaluations of slow numerical solvers to identify optimal patient-specific parameters. We address this by introducing a differentiable neural surrogate that replaces the traditional forward model. Unlike the original solver, this surrogate is fully differentiable, allowing us to solve the inverse problem using highly efficient gradient-based optimization. This approach ensures that the solution learns the biophysical constraints of tumor growth while accelerating the process by orders of magnitude. In a 3D brain tumor growth setting, our framework achieves accuracy competitive with classical optimization while reducing runtime from days to seconds. Crucially, we demonstrate that our method, though trained on synthetic data, generalizes effectively to real patient scans. These findings establish differentiable surrogates as a powerful tool for accelerating scientific machine learning in medical imaging and beyond.

Primary Subject Area: Application: Radiology

Secondary Subject Area: Interpretability and Explainable AI

Registration Requirement: Yes

Reproducibility: Will be made avaliabe upon accaptence

Visa & Travel: No

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Submission Number: 64