Task-driven assessment of experimental designs in diffusion MRI: A computational framework.

Area of Science:

• Magnetic Resonance Imaging
• Computational Neuroscience
• Medical Imaging Analysis

Background:

• Traditional computational experimental design (CED) metrics for diffusion MRI focus on parameter estimation, which may not align with task-based outcomes like clinical classification.
• Current assessment methods evaluate the fidelity of microstructural parameter recovery, not the direct task performance of the experimental design.

Purpose of the Study:

• To propose and validate a task-driven computational framework for assessing diffusion MRI experimental designs.
• To directly measure quantitative task performance using simulated experiments, moving beyond parameter-estimation accuracy.
• To demonstrate the advantages of this task-driven approach over traditional CED methods for optimizing experimental design.

Main Methods:

• Simulated diffusion MRI experiments from start to finish for a given design (protocol, model, etc.).
• Task performance computed using receiver operating characteristic (ROC) curves and area under the curve (AUC) metrics.
• Validation against clinical datasets and comparison with traditional task-agnostic assessment methods.

Main Results:

• The framework accurately predicts the qualitative form of ROC curves and the relative task performance of different designs.
• In-silico predictions of AUC correlate well with real-world clinical results.
• The proposed task-driven assessment outperforms traditional methods in identifying effective experimental designs.

Conclusions:

• The developed pipeline provides accurate, quantitative predictions of real-world task performance for diffusion MRI.
• Task-driven assessment enables the selection of experimental designs that perform better in practice.
• The framework is generalizable to various task-based quantitative MRI applications, laying groundwork for future CED frameworks.