Image denoising and model-independent parameterization for improving IVIM MRI

Variability of IVIM parameters throughout the literature is a long-standing issue, and perfusion-related parameters are difficult to interpret. We demonstrate for improving the analysis of intravoxel incoherent motion imaging (IVIM) magnetic resonance (MR) images, using image denoising and a quantitative approach that does not require imposing specific exponential models. IVIM images were acquired for 13 head-and-neck patients prior to radiotherapy. Of these, 5 patients also had post-radiotherapy scans acquired. Image quality was improved prior to parameter fitting via denoising. For this, we employed neural blind deconvolution, a method of undertaking the ill-posed mathematical problem of blind deconvolution using neural networks. The signal decay curve was then quantified in terms of area under the curve ($AUC$) parameters. Denoised images were assessed in terms of blind image quality metrics, and correlations between their derived parameters in parotid glands with radiotherapy dose levels. We assessed the method's ability to recover artificial pseudokernels which had been applied to denoised images. $AUC$ parameters were compared with the apparent diffusion coefficient ($ADC$), biexponential, and triexponential model parameters, in terms of their correlations with dose, and their relative contributions to the total variance of the dataset, obtained through singular value decomposition. Image denoising resulted in improved blind image quality metrics, and higher correlations between IVIM parameters and dose. $AUC$ parameters were more correlated with dose than traditional IVIM parameters, and captured the highest proportion of the dataset's variance. V This method of describing the signal decay curve with model-independent parameters like the $AUC$, and preprocessing images with denoising techniques, shows potential for improving reproducibility and utility of IVIM imaging.