We provide a novel method to increase SNR of segmented diffusion-weighted EPI acquisitions. Multiple gradient echoes were acquired after each diffusion-preparation and combined in an SNR-optimized way using weightings from quantitative T2* maps. The combination of diffusion-weighted echoes yielded an SNR-gain of 58% compared to single-echo dMRI data with an increase in the segmented readout duration by only 23.1 ms. The multi-echo diffusion MRI acquisition and combination were employed to acquire high-quality ex-vivo diffusion-weighted MRI data from a wild chimpanzee brain.
Typical challenges of ex-vivo diffusion MRI (dMRI) acquisitions include low diffusivity in fixed tissue requiring strong diffusion-weightings, signal drop from trapped air, image distortions and increased echo-times due to strong diffusion-weighting requirements. Highly segmented EPI (sEPI) acquisitions can be employed to counteract these challenges by shortening echo-times and reducing image distortions in ex-vivo dMRI1,2. The short sEPI readout trains can easily be repeated to acquire multiple gradient-echoes (GRE) with a negligible increase in acquisition time. However, such Multi-Echo (ME) acquisitions are not employed in dMRI, due to T2* signal decay between echoes. In functional MRI, T2* relaxation maps are employed for SNR optimal echo-combination3. In this work, an ME-sEPI Stejskal-Tanner dMRI sequence4 was developed to achieve high-quality dMRI acquisition for ex-vivo imaging of a wild chimpanzee brain on a human-scale scanner (Figure 1A). Individual echoes were combined using T2*-dependent weighting.
MRI data were acquired from the brain of a 6-year-old juvenile wild female chimpanzee from Taï National Forest (Ivory Coast). The animal died from natural cause without human interference. The brain was extracted on site by a veterinarian four hours after death and immersion-fixed with 4% paraformaldehyde. Further preparations included the removal of superficial vessels, washing out paraformaldehyde in phosphate-buffered saline and placement in Fomblin. ME-dMRI data were acquired using a 3T Connectom System (Siemens Healthineers, Erlangen, Germany) with maximum gradient strength of 300 mT/m and a flexible 23-channel surface coil5. With a total left-right brain extent of 85 mm (100 mm including container), measurement on a small-bore scanner was not possible. ME-dMRI-sEPI and ME-GRE-sEPI datasets were acquired with matched resolution and acceleration (parameters in Table 1). The ME acquisition increased the readout by only 23.1 ms, which is neglectable compared to the diffusion-preparation plus readout of a single-echo only.
MP-PCA denoising6 was employed prior to echo-combination. A T2* map was calculated by fitting an exponential model to the ME-GRE data. A voxel-wise weighting-factor $$$w_i$$$ for each echo $$$S_i$$$ was calculated based on echo-time and T2*. For optimal SNR, the individual ME-dMRI echoes were combined using weighted averaging: $$S_{comb}=\sum_{i=1}^{n} w_i S_i=\sum_{i=1}^{n}\frac{\exp{-\frac{TE(i)}{T_2^*}}}{\sum_{j=1}^{n}\exp{-\frac{TE(j)}{T_2^*}}}S_i $$
The T2*-dependence of $$$w_i$$$ yielded a voxel-specific SNR-gain compared to single-echo dMRI. To estimate the SNR-gain, the noise was approximated as Gaussian with similar variance, for all echoes. The SNR-gain was then computed as a weighted sum of Gaussian random variables.
$$\text{SNR}_{comb}=\frac{S_{comb}}{\sigma_{comb}}=\frac{S_{comb}}{\sqrt{\sum_i^nw_i^2}\sigma}$$
Deterministic diffusion tensor tracking and visualization of the processed dataset was performed using the software package brainGL [https://github.com/braingl].