Fig 1.
Summary of the proposed MRI synthetization method.
After extracting the tissue components of the multi-tissue fibre orientation maps (FOD), partial volume (PV) maps are computed and used to generate synthetic T1w and T2w contrast with Bloch equations-based simulations.
Table 1.
Magnetic properties of white matter (WM), gray matter (GM) and cerebrospinal fluid (CSF) used for the T1w and T2w signal simulations.
These magnetic properties were measured in previous studies conducted at 3T [31–33].
Fig 2.
In-vivo and synthetic T1w and T2w images obtained from the HCP-M (far left), HCP-S (middle left) and Hospital (middle and far right) datasets. The synthetic T1w images display contrasts qualitatively similar to that of the in-vivo scans for all datasets, even in presence of tumours (middle right) and post-surgical cavities (far right). The T2w contrast was similar between the synthetic and in-vivo scans for the HCP-M and HCP datasets, including the expected hyper-intensity of the cerebrospinal fluid signal. For the Hospital dataset, the in-vivo T2w images were not acquired but synthetic images provided similar contrast to that which could be expected from an in-vivo scan, including in the areas of pathology.
Table 2.
Mean brain tissue contrast measured for in-vivo and synthetic T1w scans on the Hospital, HCP-M and HCP-S datasets.
Table 3.
Mean brain tissue contrast measured for in-vivo and synthetic T2w scans on the Hospital, HCP-M and HCP-S datasets.
Fig 3.
Example of freesurfer parcellations obtained from the in-vivo (left) and synthetic (right) scans of the Hospital dataset.
The bottom row shows a magnified section of the frontal lobe. A qualitative assessment indicated that the synthetic parcellations were similar to the in-vivo parcellations in subcortical structures and near the grey-matter/white-matter interface from which tractography is typically seeded. However, the lower resolution provided by the synthetic data (2 mm isotropic) reduced the accuracy of the pial surface segmentation, as highlighted by the green arrows.
Fig 4.
Example tractography from the in-vivo and synthetic pipelines.
The rightmost column shows overlap of the in-vivo tractography (from the leftmost column; green) with the synthetic tractography (fuchsia; white indicating overlap). Tractography densities showed clear correspondences between in vivo and synthetic pipelines for HCP-S, Hospital, and HCP-M datasets. The datasets shown here were selected at random.
Fig 5.
The tractography pipeline was run first using genuine (‘in-vivo’) diffusion and structural MR images, and then run again using the same diffusion data but with either a synthetic structural image (‘synthetic’) or a structural scan acquired during another scanning session (‘in-vivo repeat’).
Top: Dice scores for overlap of tractography between runs using in-vivo data and either synthetic or in-vivo repeat data. Median dice scores did not differ significantly between synthetic and in-vivo repeat runs after correction for multiple comparisons. Bottom: Mean tract FA values sampled from the tractography, compared with the in-vivo tractography. Differences between the in-vivo versus synthetic or in-vivo-repeat runs were below the level of practical significance. Approximate percentages were calculated by dividing differences by the mean FA of all datasets (0.52). Displayed p-values are corrected for multiple comparisons.
Table 4.
Generalised Dice coefficients for 70 cortical grey matter labels (Cortical GM) or 16 deep grey-matter labels (Deep GM).