Moving Beyond Single-Model Assumptions in DCE-MRI: Probabilistic Nested Model Selection for Enhanced Tumor Microvasculature Analysis

Motivation: The current nested model selection (NMS) method for pharmacokinetic analysis of Dynamic Contrast Enhanced (DCE) MRI data assumes a single physiological model per voxel, potentially overlooking mixed-model scenarios in complex tissues like tumors. Goal(s): To develop a probabilistic NMS (PNMS) technique that estimates voxel-wise probabilities of multiple physiological models for more accurate microvasculature parameter estimation in DCE-MRI. Approach: Using DCE-MRI data from a rat glioblastoma model, we employed a Kohonen Self-Organizing Map (K-SOM) to estimate model probabilities and compared them to conventional NMS technique. Results: PNMS provided comparable accuracy, with faster computation and more reliable permeability estimates. Impact: Probabilistic nested model selection has the potential to improve the accuracy of DCE-based microvasculature parameter estimation. It enhanced DCE-MRI based tumor microenvironmental information assessment, that may translate to more precise diagnoses, targeted therapies, and informed clinical decision-making for cancer patients.