Model-free denoising of multi b-value diffusion-weighted MR images using principal component analysis: simulations and in vivo results

We present a principal component analysis (PCA) toolkit for mode-free denoising of multi b-value diffusion-weighted images for clinical use. In simulations, PCA-denoising suppressed the random noise equally well (up to 55%) as synthetic MRI. Contrary to synthetic MRI (systematic error up to 29% of total signal intensity), PCA-denoising did not introduce any systematic errors (<2%). In volunteer and patient image data, PCA-denoising resulted in sharper and less noisy images than synthetic MRI, which resulted in sharper and clearer tumour boundaries. In conclusion, our PCA-denoising toolkit is promising for denoising b-value images for clinical use.