On the battle between Rician noise and phase-interferences in DWI

Introduction: It has been shown that complex averaging benefits diffusion-weighted imaging (DWI) by reducing the influence of the noise bias in the ADC and DTI (1). Magnitude averaging in DWI is traditionally used to combine multi-directional and multiNEX diffusion data – to avoid the complication of random phase offsets, due to motion that occurs during the DW gradients. However, magnitude averaging introduces a non-zero bias of the signal expectation value (a.k.a. Rician noise (2-4)) and underestimates the calculated ADC (1,5). This bias grows with lower SNR in each diffusion weighted image – and this is a particular concern for high-resolution, thin-slice DWI/DTI. Even in the clinical practice – where the non-quantitative mean DWI (a.k.a. isotropic DWI) is of most importance – the hazy looking image can confound image interpretation and diagnostic confidence. The advent of phase correction techniques – such as the triangular windowing approach initially proposed for PROPELLER DWI data [6] – has been shown to be a fast and effective phase correction approach for navigated DW EPI data. Here, each complex DW k-space is reconstructed twice into two temporary images, one of them after the application of a triangular window function of radius r in k-space. The non-windowed image is then subtracted with the phase information content in the corresponding windowed image (Fig. 1). The radius of the triangular window will determine how much phase is removed. While a large r will help to remove image artifacts due to pulsatile brain motion that occurs during the DW gradients, it will also ‘approach magnitude averaging’ or in other words – result in pronounced Rician noise in the mean DWI data after averaging the multiple NEX, repetitions, and diffusion directions together. In this abstract, we report on the best choice of triangular windowing radius for use in clinical practice. Over 1,200 patients have been scanned at our hospital with our GRAPPA-accelerated EPI sequence [7] and online reconstruction developed inhouse. With this data, we have determined the most suitable triangular window radius that minimizes the amount of Rician noise in the final iso-DWI data, without introducing phase cancellations.