Image denoising and model-independent parameterization for IVIM MRI

Area of Science:

• Magnetic Resonance Imaging
• Medical Physics
• Radiotherapy

Background:

• Intravoxel incoherent motion (IVIM) imaging is crucial for assessing tissue microcirculation.
• Current IVIM parameterization faces challenges with high variability and interpretability.
• Improving IVIM image quality and parameterization is essential for clinical applications.

Purpose of the Study:

• To enhance intravoxel incoherent motion (IVIM) magnetic resonance imaging (MRI) quality.
• To introduce a novel image denoising technique using neural blind deconvolution.
• To utilize model-independent parameterization (AUC) for IVIM signal quantification.

Main Methods:

• Acquired IVIM MRI scans for head-and-neck cancer patients before and after radiotherapy.
• Applied neural blind deconvolution for image denoising.
• Quantified the IVIM signal decay curve using area under the curve (AUC) parameters.
• Assessed image quality using blind metrics, total variation (TV), and contrast-to-noise ratio (CNR).

Main Results:

• Denoising significantly improved image quality metrics and smoothed the signal decay curve.
• Denoising reduced image total variation (TV) and generally increased parameter contrast-to-noise ratios (CNRs).
• AUC parameters demonstrated stronger correlations with radiotherapy dose levels and higher relative importance compared to exponential models.

Conclusions:

• The proposed denoising technique and AUC parameterization enhance IVIM MRI quality and reproducibility.
• Model-independent AUC parameters offer a more robust approach to IVIM data analysis.
• This method holds potential for improved functional utility and clinical translation of IVIM imaging.