Deep learning-based diffusion tensor image generation model: a proof-of-concept study

This study developed an image-to-image translation model for synthesizing diffusion tensor images (DTI) from conventional diffusion weighted images. Thirty-two healthy volunteers participated, providing DTI and DWI data with six and three motion probing gradient (MPG) directions, respectively. The image-to-image translation model was applied six times for each MPG direction. Regions of interest (ROIs) in specific brain areas were created and applied to compare mean values and signal-to-noise ratio (SNR) between original and synthetic DTI maps. The Bland–Altman plot demonstrated similar distributions despite slightly lower SNR in the synthetic data. Synthetic DTI can be generated effectively from conventional DWI using this model.

Source Nature

Author: Neurologica