NIMG-17. TOWARDS IMAGE - BASED PERSONALIZATION OF GLIOBLASTOMA THERAPY: A COMPREHENSIVE VALIDATION STUDY OF A NOVEL, DEEP LEARNING - DRIVEN TUMOR GROWTH MODEL FOR INDIVIDUALIZED RADIOTHERAPY

Marie‐Christin Metz, Michael Griessmair, Ivan Ezhov, Jan C. Peeken, Jonas Weidner, Bjoern Menze, Benedikt WiestlerNeuro-Oncology 2023 · DOI

Introduction

Glioblastoma (GBM) remains one of the most challenging malignancies to treat due to its highly invasive and diffuse growth pattern. Unlike many solid tumors, GBM cells infiltrate the surrounding brain parenchyma far beyond the visible borders on standard magnetic resonance imaging (MRI). Current radiotherapy protocols typically address this by applying a uniform margin around the visible tumor to create a Clinical Target Volume (CTV). However, this "one-size-fits-all" approach often fails to account for the unique infiltration patterns of individual tumors, potentially leading to either under-treatment of high-risk areas or unnecessary irradiation of healthy tissue.

Computational tumor growth modeling offers a potential solution by estimating patient-specific tumor cell density across the entire brain. This study validates a novel growth model that combines classical mathematical frameworks with deep learning to personalize radiotherapy planning.

Methods and Materials

The study utilized a Fisher-Kolmogorov (FK) growth model, which characterizes tumor dynamics through two primary parameters:

  1. Infiltration ($D_w$): The rate at which tumor cells diffuse through brain tissue.
  2. Proliferation ($\rho$): The rate of tumor cell multiplication.

A deep learning-driven approach was employed to adjust the FK model to the specific anatomy of each patient. Validation was conducted across three distinct cohorts:

  • The Cancer Genome Archive (TCGA) (n=124): Used to correlate model parameters with genetic pathway activation maps (generated via PARADIGM) and overall survival.
  • UCSF Glioma Dataset (n=397): Used for large-scale validation of the correlation between model parameters and clinical outcomes.
  • Institutional Dataset (n=30): Used to compare model-derived target volumes against standard clinical radiotherapy plans, specifically focusing on recurrence coverage.

Results

Biological Plausibility

A critical finding of this study was the biological validation of the computed parameters. The estimated proliferation parameter ($\rho$) showed a significant correlation with 11 genetic proliferation pathways. This suggests that the deep learning-enhanced model successfully captures the underlying biological behavior of the tumor rather than simply performing image-based curve fitting.

Clinical and Prognostic Significance

The study identified the $D_w/\rho$ ratio as a significant prognostic marker. Both this ratio and the simulated tumor volume (defined at a 0.5 tumor cell density threshold) were significantly inversely correlated with overall survival ($p < 0.05$) in both the TCGA and UCSF cohorts.

Radiotherapy Optimization

In the institutional cohort, researchers compared standard clinical target volumes with model-derived volumes. Using a cutoff value of 0.5 tumor cell density to define the target volume yielded the most favorable results. This approach significantly improved the coverage of future tumor recurrences without a significant increase in the total radiation volume compared to standard plans.

Discussion and Conclusion

The integration of deep learning with mathematical growth models represents a significant step toward individualized glioblastoma therapy. By demonstrating that computed growth parameters correlate with both genetic pathways and clinical survival, this study provides a robust validation of the model's biological and clinical relevance.

The ability to improve recurrence coverage without increasing the total radiation dose suggests that model-guided radiotherapy could enhance treatment efficacy while maintaining safety. While these retrospective results are promising, prospective clinical trials are necessary to determine if implementing these personalized plans leads to improved patient survival. Future work will also focus on integrating these computational pipelines directly into clinical treatment planning systems to streamline the workflow for radiation oncologists.

By bridging the gap between mathematical modeling and clinical practice, this research moves us closer to a future where glioblastoma treatment is precisely tailored to the unique anatomical and biological profile of every patient.

Summary updated

Abstract

Abstract BACKGROUND The diffuse growth pattern of glioblastoma is one of the main challenges for accurate treatment planning. Computational tumor growth modeling has emerged as a promising tool to guide personalized therapy by estimating the tumor cell density in the whole brain. Here, we performed biological and clinical validation of a novel growth model, aiming to close the gap between the experimental state and clinical implementation. METHODS 124 patients from The Cancer Genome Archive (TCGA) and 397 patients from the UCSF Glioma Dataset were assessed for significant correlations between clinical data, genetic pathway activation maps (generated with PARADIGM; TCGA only), and infiltration (Dw) as well as proliferation (ρ) parameters stemming from a Fisher-Kolmogorov growth model that is adjusted to the patient’s anatomy using deep learning. To further evaluate clinical potential, we performed the same growth modeling on preoperative MRI data from 30 patients of our institution and compared model-derived tumor volumes (that serve as alternative clinical target volumes) and recurrence coverage with standard radiotherapy plans. RESULTS The parameter ratio Dw/ρ (p &amp;lt; 0.05 in TCGA) as well as the simulated tumor volume (0.5 tumor cell density) (p &amp;lt; 0.05 in TCGA/UCSF) were significantly inversely correlated with overall survival. We found a significant correlation between 11 proliferation pathways and the estimated proliferation parameter ρ, proving its biological plausibility. Depending on the cutoff value for tumor cell density, we observed significant improvement of recurrence coverage without significantly increased total radiation volume utilizing model-derived target volumes instead of standard radiation plans. More precisely, a cutoff value of 0.5 tumor cell density yielded the best results here. CONCLUSIONS Identifying a significant correlation between computed growth parameters, and clinical and biological data, we highlight the potential of tumor growth modeling for individualized therapy of glioblastoma. This might improve accuracy of radiation planning in the near future.
BibTeX
@inproceedings{metz2023nimg,
  author = {Marie‐Christin Metz and Michael Griessmair and Ivan Ezhov and Jan C. Peeken and Jonas Weidner and Bjoern Menze and Benedikt Wiestler},
  title = {{NIMG-17. TOWARDS IMAGE - BASED PERSONALIZATION OF GLIOBLASTOMA THERAPY: A COMPREHENSIVE VALIDATION STUDY OF A NOVEL, DEEP LEARNING - DRIVEN TUMOR GROWTH MODEL FOR INDIVIDUALIZED RADIOTHERAPY}},
  booktitle = {Neuro-Oncology},
  year = {2023},
  doi = {10.1093/neuonc/noad179.0713},
}
Figure