Towards Virtual MR Imaging: Predicting Diffusion-Weighted Brain MR Images from T2-Weighted Images Using Convolutional Neural Networks

Oral Presentation at the European Congress of Radiology, Vienna, 2019


In a quest for standardisation and speeding-up for MR Imaging, there is a move towards having a 'universal' MRI sequence from which all MR contrasts can be obtained. We propose first-of-its-kind Virtual MR Imaging where we predict "Virtual" Diffusion-Weighted (DW) images of the brain from axial T2W images using Convolutional Neural Networks (CNN).

Methods and Materials

100 whole brain MRI scans of patients with no abnormality and 30 with acute infarcts, comprising of 25 T2W and DWI (b=1000) slices each, acquired retrospectively, were fed into a 15 layered CNN model with a 75-25 training-validation split. T2W images were assigned as the input to predict DW images. Binary Cross Entropy was used as the loss function and training took 48 hours on an Nvidia 1080ti GPU. For testing T2W images from 5 independent cases, having 7 T2 hyperintensities with corresponding diffusion restriction (group-1) and 16 with no diffusion restriction (group-2), were also analysed.


Binary Cross Entropy of 0.15 for normal and 0.11 for infarct cases was obtained. The model took 750 ms to produce each image. In the test cases, 6 out of 7 T2 hyperintensities in group-1 showed diffusion restriction and all 16 T2 hyperintensities in group-2 did not show restriction. The one 'missed' lesion in group-1 was 2.5 mm diameter. Blurring of virtual DW images was observed, which did not impede clinical judgement.


We demonstrate a novel use-case for Deep Learning in reducing the MRI exam time and potentially creating a single universal MRI sequence.