Accelerated multi-shot diffusion MRI using deep learning denoising

Multi-shot EPI is commonly used to compensate for geometric distortions and increase spatial resolution in body diffusion MRI, with a price tag of longer scan times. This work presents an alternative technique to k-space undersampling to accelerate the acquisition, which is based on reducing the number of repetitions at high b-value and denoising the resulting images using a convolutional neural network. The proposed deep learning denoising technique is demonstrated to accelerate the acquisition of multi-shot diffusion MRI acquisition of patients with rectal cancer and reduce the scan time beyond the duration of a single-shot diffusion MRI acquisition.