Principal component analysis for model-free denoising of multi b-value diffusion-weighted images

We introduce principal component analyses (PCA) as a denoising technique for diffusion-weighted MRI (DWI) that is independent of the diffusion attenuation model. PCA denoises DWI data using only informative components while removing noisy ones. We show that it outperforms model-based denoising in simulations as well as in vivo. In simulations, PCA-denoising resulted in smaller systematic errors, while random errors were similar. In vivo, PCA-denoising rendered less noisy images and when motion was present, PCA recovered certain structures that were obscured by motion in model-based denoising. In conclusion, PCA-denoising is a powerful model-free tool for denoising DWI data.