Radiomic Deformation and Textural Heterogeneity (R-DepTH) Descriptor to Characterize Tumor Field Effect: Application to Survival Prediction in Glioblastoma
Marwa Ismail, Prateek Prasanna, Kaustav Bera, Volodymyr Statsevych, Virginia B. Hill, Gagandeep Singh, Sasan Partovi, Niha G. Beig, Sean D. McGarry, Peter S. LaViolette, Manmeet Ahluwalia, Anant Madabhushi, Pallavi Tiwari
Glioblastoma (GBM) is a highly aggressive brain tumor where the "tumor field effect"—the impact of cancer beyond visible margins—plays a critical role in patient outcomes. This study introduces the Radiomic-Deformation and Textural Heterogeneity (r-DepTH) descriptor, an MRI-based prognostic marker that combines biomechanical tissue deformation in the normal parenchyma with local morphological heterogeneity within the tumor. Evaluated on 207 GBM cases across three multi-institutional cohorts, r-DepTH demonstrated superior survival risk-stratification compared to standard radiomic and deep learning approaches, achieving high hazard ratios and robust performance across different clinical sites.
Introduction
Glioblastoma is traditionally assessed based on the visible tumor mass on MRI. However, cancer is a systemic disease that exerts a "tumor field effect" well beyond these margins. Two critical but often unquantified factors are the mass effect—where tumor growth physically deforms healthy brain structures—and invisible infiltration of cancer cells into surrounding tissue. Standard radiomics often neglects these biomechanical "ripples." This paper addresses this gap by quantifying the physical and morphological impact of the tumor on the entire brain environment.
Methodology
The proposed r-DepTH descriptor integrates two primary components to capture the tumor field effect:
Biomechanical Deformation: Using diffeomorphic registration (specifically the ANTs SyN toolbox), patient MRI scans are aligned to a healthy brain atlas. The resulting inverse mapping quantifies how much healthy tissue is displaced by the tumor. These deformations are measured in 12 concentric rings extending up to 60mm from the tumor margin.
Textural Heterogeneity (COLLAGE): The Co-occurrence of Local Anisotropic Gradient Orientations (COLLAGE) descriptor is applied within the tumor confines to capture morphological disorganization and infiltration patterns.
Overview of the r-DepTH workflow, integrating tumor segmentation, deformation field computation, and textural heterogeneity mapping for survival analysis.
A LASSO (Least Absolute Shrinkage and Selection Operator) model selects the most predictive features from a pool of 450 to generate a survival risk score.
Experimental Design
The study utilized a total of 207 GBM cases across three distinct cohorts to ensure the model's generalizability:
Cohort 1 (Cleveland Clinic - CCF): 53 cases.
Cohort 2 (The Cancer Imaging Archive - TCIA): 75 cases.
Cohort 3 (Medical College of Wisconsin - MCW): 79 cases.
The researchers used a "round-robin" training approach: training on one cohort and testing independently on the other two. This rigorous validation ensures the model captures fundamental biological truths rather than site-specific imaging artifacts.
Results
r-DepTH successfully stratified patients into high- and low-risk groups with high statistical significance ($p < 0.0001$ in several tests).
Predictive Power: The model achieved Concordance indices (C-indices) ranging from 0.60 to 0.75 across different test sets.
Hazard Ratios (HR): High-risk patients identified by r-DepTH were significantly more likely to have shorter survival. For instance, when training on Cohort 3, the model yielded HRs of 24 (CI 10–57) and 12 (6–21) on Cohorts 1 and 2, respectively.
Comparative Performance: As shown in Figure 5, r-DepTH consistently outperformed standard radiomics, deep features, and individual components (COLLAGE or deformation alone) across all cohorts.
Kaplan-Meier survival curves comparing r-DepTH against standard radiomics, deep features, COLLAGE, and deformation alone across three independent cohorts.
The findings highlight that the "normal" brain tissue surrounding a tumor contains vital prognostic information. Specifically, deformations up to 60mm away serve as a proxy for tumor aggressiveness. The robustness of r-DepTH across different institutions suggests it captures fundamental biological truths. Future work will explore the directionality of deformations and integrate molecular markers (e.g., IDH or MGMT status) to further refine prognostic accuracy and guide treatment decisions in a prospective clinical setting.
Summary updated
Abstract
The concept of tumor field effect implies that cancer is a systemic disease with its impact way beyond the visible tumor confines. For instance, in Glioblastoma (GBM), an aggressive brain tumor, the increase in intracranial pressure due to tumor burden often leads to brain herniation and poor outcomes. Our work is based on the rationale that highly aggressive tumors tend to grow uncontrollably, leading to pronounced biomechanical tissue deformations in the normal parenchyma, which when combined with local morphological differences in the tumor confines on MRI scans, will comprehensively capture tumor field effect. Specifically, we present an integrated MRI-based descriptor, radiomic-Deformation and Textural Heterogeneity (r-DepTH). This descriptor comprises measurements of the subtle perturbations in tissue deformations throughout the surrounding normal parenchyma due to mass effect. This involves non-rigidly aligning the patients’ MRI scans to a healthy atlas via diffeomorphic registration. The resulting inverse mapping is used to obtain the deformation field magnitudes in the normal parenchyma. These measurements are then combined with a 3D texture descriptor, Co-occurrence of Local Anisotropic Gradient Orientations (COLLAGE), which captures the morphological heterogeneity and infiltration within the tumor confines, on MRI scans. In this work, we extensively evaluated r-DepTH for survival risk-stratification on a total of 207 GBM cases from 3 different cohorts (Cohort 1 ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${n}{1} = {53}$ </tex-math></inline-formula> ), Cohort 2 ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${n}{2} = {75}$ </tex-math></inline-formula> ), and Cohort 3 ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${n}{3} = {79}$ </tex-math></inline-formula> )), where each of these three cohorts was used as a training set for our model separately, and the other two cohorts were used for testing, independently, for each training experiment. When employing Cohort 1 for training, r-DepTH yielded Concordance indices (C-indices) of 0.7 and 0.65, hazard ratios (HR) and Confidence Intervals (CI) of 10 (6 − 19) and 5 (3 − 8) on Cohorts 2 and 3, respectively. Similarly, training on Cohort 2 yielded C-indices of 0.6 and 0.7, HR and CI of 1 (0.7 − 2) and 3 (2 − 5) on Cohorts 1 and 3, respectively. Finally, training on Cohort 3 yielded C-indices of 0.75 and 0.63, HR and CI of 24 (10 − 57) and 12 (6 − 21) on Cohorts 1 and 2, respectively. Our results show that r-DepTH descriptor may serve as a comprehensive and a robust MRI-based prognostic marker of disease aggressiveness and survival in solid tumors.
BibTeX
@article{ismail2022radiomic,
author = {Marwa Ismail and Prateek Prasanna and Kaustav Bera and Volodymyr Statsevych and Virginia B. Hill and Gagandeep Singh and Sasan Partovi and Niha G. Beig and Sean D. McGarry and Peter S. LaViolette and Manmeet Ahluwalia and Anant Madabhushi and Pallavi Tiwari},
title = {{Radiomic Deformation and Textural Heterogeneity (R-DepTH) Descriptor to Characterize Tumor Field Effect: Application to Survival Prediction in Glioblastoma}},
journal = {IEEE Trans. Medical Imaging},
year = {2022},
url = {https://openreview.net/forum?id=BOEy9X52Fi},
}