Supervised denoising of diffusion-weighted magnetic resonance images using a convolutional neural network and transfer learning

In this paper, we propose a method for reducing thermal noise in diffusion-weighted magnetic resonance images (DWI MRI) of the brain using convolutional neural network (CNN) trained on realistic, synthetic MR data. Two reference methods are considered: a) averaging repeated scans, widespread used clinics to improve signal-to-noise ratio and b) blockwise Non-Local Means (NLM) filter, one post-processing frequently DWI denoising. To obtain training data transfer learning, effects echo-planar imaging (EPI) – Nyquist ghosting ramp sampling modelled data-driven fashion. These introduced digital phantom anatomy (BrainWeb). Real maps obtained from MRI scanner with brain-DWI-designed protocol later combined simulated, noise-free EPI images. The Point Spread Function is measured DW image an AJR-approved geometrical phantom. Inter-scan patient movement captured scan healthy volunteer registration. denoising applied simulated real brain. quality denoised evaluated at several ratios. characteristics residuals studied thoroughly. A diffusion investigate influence ADC measurements. also GRAPPA dataset. We show that our outperforms NLM allows significant reduction time by lowering number scans. analyse CNN denoisers point out challenges accompanying method.