R-Index: A Robust Metric for IVIM Parameter Estimation on Clinical MRI Scanners

TL;DR

This paper introduces the R-index, a new robust metric for IVIM parameter estimation in clinical MRI that reduces collinearity and improves reproducibility under low SNR conditions.

Contribution

The study proposes the R-index, a novel metric that enhances IVIM parameter estimation robustness by mitigating collinearity and reducing uncertainty in clinical MRI settings.

Findings

The R-index significantly reduces estimation deviation compared to traditional IVIM parameters.

Parameter collinearity was confirmed with 32% of voxels showing significant correlations.

The R-index improves reproducibility of IVIM measurements at low SNR.

Abstract

Background: Intravoxel Incoherent Motion (IVIM) model characterizes both water diffusion and perfusion in tissues, providing quantitative biomarkers valuable for tumor tissue characterization. However, parameter estimation based on this model is challenging due to its ill-posed nature, resulting in poor reproducibility, particularly at low signal to noise ratios (SNRs) in a clinic scenario. Purpose: This study analyzes the uncertainty of IVIM model fitting, quantifies parameter collinearity, and introduces a new index with enhanced robustness to enhance clinical applicability of the IVIM model. Study Type: Prospective. Population: One healthy volunteer. Field Strength/Sequence: 1.5T; single-shot EPI DWI. Assessment: The probability distributions of estimated IVIM parameters were evaluated across a clinically relevant range. Collinearity among parameters was assessed and a new metric, the R-index, was proposed. The R-index linearly combines individual IVIM parameters to mitigate collinearity and reduce estimation uncertainty. Simulation and a volunteer study was conducted to validate the presence of parameter collinearity and to assess the robustness of the R-index. Statistical Tests: N/A Results: In simulation studies with a typical clinical setting (SNR = 20), normalized IVIM parameters exhibited mean standard deviations ranging from 0.107 to 0.269, while the R-index showed a reduced deviation of 0.064. Repeated scans in a healthy volunteer confirmed the presence of parameter collinearity, with 32% of voxels exhibiting statistically significant correlations (p < 0.05) among fitted IVIM parameters, and a mean Pearson correlation coefficient of r = -0.96. Data Conclusion: The R-index provides a robust metric for IVIM model fitting under low SNR conditions typical of clinical MRI, offering improved reproducibility and potential for broader clinical applicability.