Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TumorFlow: Physics-Guided Longitudinal MRI Synthesis of Glioblastoma Growth

Valentin Biller, Niklas Bubeck, Lucas Zimmer, Ayhan Can Erdur, Sandeep Nagar, Anke Meyer-Baese, Daniel Rückert, Benedikt Wiestler, Jonas Weidner

TL;DR

A biophysically-conditioned generative framework that synthesizes biologically realistic 3D brain MRI volumes from estimated, spatially continuous tumor-concentration fields and can generate temporally coherent sequences with realistic changes in tumor appearance and surrounding tissue response is presented.

Abstract

Glioblastoma exhibits diverse, infiltrative, and patient-specific growth patterns that are only partially visible on routine MRI, making it difficult to reliably assess true tumor extent and personalize treatment planning and follow-up. We present a biophysically-conditioned generative framework that synthesizes biologically realistic 3D brain MRI volumes from estimated, spatially continuous tumor-concentration fields. Our approach combines a generative model with tumor-infiltration maps that can be propagated through time using a biophysical growth model, enabling fine-grained control over tumor shape and growth while preserving patient anatomy. This enables us to synthesize consistent tumor growth trajectories directly in the space of real patients, providing interpretable, controllable estimation of tumor infiltration and progression beyond what is explicitly observed in imaging. We evaluate the framework on longitudinal glioblastoma cases and demonstrate that it can generate temporally coherent sequences with realistic changes in tumor appearance and surrounding tissue response. These results suggest that integrating mechanistic tumor growth priors with modern generative modeling can provide a practical tool for patient-specific progression visualization and for generating controlled synthetic data to support downstream neuro-oncology workflows. In longitudinal extrapolation, we achieve a consistent 75% Dice overlap with the biophysical model while maintaining a constant PSNR of 25 in the surrounding tissue. Our code is available at: https://github.com/valentin-biller/lgm.git

TumorFlow: Physics-Guided Longitudinal MRI Synthesis of Glioblastoma Growth

TL;DR

A biophysically-conditioned generative framework that synthesizes biologically realistic 3D brain MRI volumes from estimated, spatially continuous tumor-concentration fields and can generate temporally coherent sequences with realistic changes in tumor appearance and surrounding tissue response is presented.

Abstract

Glioblastoma exhibits diverse, infiltrative, and patient-specific growth patterns that are only partially visible on routine MRI, making it difficult to reliably assess true tumor extent and personalize treatment planning and follow-up. We present a biophysically-conditioned generative framework that synthesizes biologically realistic 3D brain MRI volumes from estimated, spatially continuous tumor-concentration fields. Our approach combines a generative model with tumor-infiltration maps that can be propagated through time using a biophysical growth model, enabling fine-grained control over tumor shape and growth while preserving patient anatomy. This enables us to synthesize consistent tumor growth trajectories directly in the space of real patients, providing interpretable, controllable estimation of tumor infiltration and progression beyond what is explicitly observed in imaging. We evaluate the framework on longitudinal glioblastoma cases and demonstrate that it can generate temporally coherent sequences with realistic changes in tumor appearance and surrounding tissue response. These results suggest that integrating mechanistic tumor growth priors with modern generative modeling can provide a practical tool for patient-specific progression visualization and for generating controlled synthetic data to support downstream neuro-oncology workflows. In longitudinal extrapolation, we achieve a consistent 75% Dice overlap with the biophysical model while maintaining a constant PSNR of 25 in the surrounding tissue. Our code is available at: https://github.com/valentin-biller/lgm.git
Paper Structure (4 sections, 3 equations, 5 figures, 1 table)

This paper contains 4 sections, 3 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. Method
  3. Experiments
  4. Conclusion

Figures (5)

  • Figure 1: Representative longitudinal tumor growth prediction. Starting from the preoperative state (week 0), a biophysical growth model is used to simulate time-resolved tumor concentration fields (blue), which serve as conditioning for image synthesis. The model generates T1c, FLAIR, T1, T2 modalities (green), preserving anatomy while capturing modality-specific progression. A standard segmentation model applied to the generated images yields time-consistent masks (orange).
  • Figure 2: Architecture of TumorFlow. Modality, tissue segmentations, and tumor concentrations are provided as conditioning inputs to the latent generative model. Orange boxes denote trainable modules, whereas blue boxes indicate frozen, pre-trained components. During inference, a full growth trajectory can be extrapolated from a biophysical simulation.
  • Figure 3: Qualitative Comparison. Reconstructions from different methods.
  • Figure 4: Left: Tumor Consistency. Dice across time steps, assessing tumor growth consistency. Right: Tissue Consistency. PSNR across time steps measuring structural drift in healthy tissue. Stable values indicate coherent generation.
  • Figure 5: Left: Real patients. Comparing TumorFlow and groundtruth tumor development in two patients from Lumiere suter2022lumiere. The red arrow points to a new tumor lesion in the corpus callosum. Right: Effect of corruption initialization on longitudinal generation. Top: Dice between generated tumors and conditioning. Bottom: PSNR between consecutive non-tumorous regions.