For the ex vivo studies, data were those from a previous study [11]. Eight rhesus monkey brains were perfusion-fixed with 4% formalin for 2-3 days as reported in this study. The sample was rinsed with water to remove formalin and immersed in water containing 0.018 mg/ml MION particles. Submersion in aqueous solution prevented susceptibility artifacts at the tissue-air interface. Images suffered from Gibbs ringing artifact due to the strong signal that was detected if pure water or saline was used. Addition of MION particles reduced the T2* of the solution, thereby eliminating Gibbs ringing. MION particles were large enough to remain outside the brain tissue. The sample was immobilized in the container using plastic holders.

For the in vivo studies, six rhesus monkeys (2 male, 4 female) were anesthetized initially with telazol (3-5 mg/kg) and maintained with 1-1.5% isoflurane during MRI scans as per institutional approval. The eyes were coated with PuraLube to prevent dryness. Lactate ringer solution (10 ml/kg/hr) was administered via an i.v. line. The animals were placed in a holder using a mouth-piece and ear bars. Heart-rate, respiration, temperature and blood pressure were continuously monitored during the experiment. All procedures were performed with Institutional Animal Care and Use Committee approval.

All experiments were done on a Siemens® TRIO 3 Tesla (Siemens, Erlangen, Germany) human scanner. Stimulated Echo Acquisition Mode (STEAM) [20,21] sequence was used to allow an increase in diffusion by increasing the mixing time, TM. Geometric averaging was used for STEAM to reduce the contribution of cross terms [22, 23].

The ex vivo MRI parameters were: 16-shot STEAM EPI, TR/TE = 3000 ms/80 ms, δ = 17 ms, Δ = 21 ms for short t_diff (48 ms) and 169 ms for long t_diff (192 ms), directions = 30, b = 1700 s/mm², resolution = 0.54 x 0.54 x 2 mm and number of averages = 34 for short t_diff, 70 for long t_diff. These sequences are referred to as STEAM48 and STEAM196 henceforth.

The in vivo MRI parameters were: 4-shot EPI, TR /TE = 3000/63 ms, δ = 17 ms, Δ = 31 ms for short t_diff (50 ms) and 181 ms for long t_diff (200 ms), directions = 60, b = 1200 s/mm², resolution = 1x1x1 mm and number of averages = 3 for short t_diff, 6 for long t_diff. These in vivo sequences are referred to as STEAM50 and STEAM200.

DSE was also acquired with MRI parameters similar to the STEAM50 except TE = 91 ms which was the shortest TE possible. The numbers of averages among STEAM DTI acquisitions at short and long t_diff and DSE were adjusted to match SNR for comparison.

In addition to the commonly used DTI parameters such as ADC and FA, geometric measures of anisotropy – CL, CP and CS [11, 16] – were also analyzed to evaluate the diffusion sensitivity at long t_diff. CL, CP and CS are measures of linear, planar and spherical anisotropy and are derived from the principle eigen values λ₁, λ₂ and λ₃ as follows [16]:

$\mathrm{CS}=\frac{{}_3}{3\left(\right)},\mathrm{CP}=\frac{2\left({}_2{}_3\right)}{3\left(\right)},\mathrm{CL}=\frac{\left({}_1{}_2\right)}{3\left(\right)},\mathrm{where}\left(\right)=\frac{{}_1+{}_2+{}_3.}{3}$ Geometric measures were plotted in the form of three phase plots [11]. To that effect, x and y co-ordinates [32] to represent anisotropy can be plotted using

Changes in anisotropic diffusion in terms of CL and CP and isotropic diffusion in term of CS due to long t_diff can be depicted on these plots.

Further, to verify the coherence of tensor directionality along a fiber tract, directional entropy (DE) [17] was also measured. DE is inversely proportional to the inter-voxel tensor coherence along a nerve fiber. DTI at long t_diff would produce more coherent tensor orientations along a fiber tract and hence a reduced DE is expected.

FA, ADC, parallel and perpendicular diffusivity were calculated. Nerve fiber tracks were tracked from the tensor data, using either line propagation techniques or probabilistic maps. While line propagation techniques such as Fiber Assignments by Continuous Tracking (FACT) [24] offer a comparatively simple approach to visualize various neural pathways, the technique relies heavily on the estimated tensor data and hence on image quality and SNR. Probabilistic approaches [25-27] compute a 3D connectivity map representing all the likely neural pathways that may exist at the selected seed in the brain volume. However, these techniques require a priori knowledge of the neural tracts in the brain region of interest were a database of numerous DTI scans is required to obtain comprehensive information about the similarities and differences in each nerve fiber tract across different animals of the same species. In this study, tractography was performed using the Diffusion TENSOR Visualizer [28], a software package that uses an advanced line propagation method.

Fig. (2) shows the ADC and FA values in WM and GM for the ex vivo and in vivo studies using the typical DSE sequence. The ADC values in the fixed tissue were much lower than those in the in vivo data in the white matter WM (p < 0.01) (Fig.2a). ADC was also measured in the fixed tissue and found to be 0.27 ± 0.02 x 10^-3 mm²/s in the grey matter as opposed to 0.74 ± 0.01 x 10^-3 mm²/s in vivo. The WM ADC in fixed tissue was 0.16 ± 0.02 x 10^-3 mm²/s and in vivo, it was 0.81 ± 0.02 x 10^-3 mm²/s. Surprisingly, WM ADC was higher than GM ADC, which also has been reported previously in some WM regions [29]. In addition, the ratio of parallel diffusivity to perpendicular diffusivity [i.e., 2λ₁/ (λ₂+ λ₃)] was 1.53 ± 0.08 for ex vivo WM and 2.10 ± 0.08 for in vivo WM.

Fig. (2). (a) ADC in the GM and WM in the in vivo and ex vivo experiments. Compared to the in vivo values, ADC in the WM was reduced by almost by 61 ± 5% and ADC in the GM decreased by 39 ± 8% in the ex vivo samples. (b) Comparison of FA in the in vivo and ex vivo samples. FA reduced by almost 28% in the WM in the ex vivo study while in the GM the decrease was less prominent at 10% compared to the in vivo FA. (Ex vivo data obtained from [11])

The FA values in the fixed tissue were also much lower than those in the in vivo data in the white matter WM (p < 0.01) (Fig.2b). FA decreased by 28% in fixed WM compared to in vivo WM. FA decrease in GM was much less (10%).

The ex vivo and in vivo WM and GM FA values at the two t_diff are summarized in Table 1.Ex vivo GM FA was unchanged, whereas ex vivo WM FA increased by 11.5 ± 5.0% (p < 0.05) for t_diff = 196 ms compared to t_diff = 48 ms. Similarly, in vivo GM FA was unchanged, whereas in vivo WM FA increased by 6.03 ± 0.52% for t_diff = 200 ms compared to t_diff = 50 ms (p<0.01). Further, FA estimated with STEAM50 and DSE in the in vivo studies were not significantly different. Similar results were observed for the ex vivo study, thereby confirming that the DTI parameters evaluated with STEAM and DSE sequences do not differ significantly. SNRs calculated for the STEAM50, STEAM200 and DSE sequence were 24.58, 22.87 and 28.45 respectively.

Three-phase plots were obtained for the in vivo experiments at t_diff = 50 ms and t_diff = 200 ms (Fig.3a). The three-phase plot shows voxels of the cortical terminations of the callosal fiber tract as shown in the inset. The tendency for linear anisotropic diffusion increased significantly (p < 0.05) at t_diff = 200 ms as opposed to t_diff = 50 ms. Similar results were also obtained with DSE sequence. Fig. (3b) compares the geometric measures of diffusion for DSE and STEAM200 in the cortico-spinal tract. DE along the same callosal fibers (Table2) were reduced significantly (p<0.01). Similar results were also obtained for the cortico-spinal tract. Thus the tensors along the nerve fiber pathway not only became more linearly anisotropic (i.e. directional) but also showed increased inter-voxel coherence. Comparison of the directional entropy revealed that DE decreased by 17 ± 4% in the callosum and by 10 ± 2% in the cortico-spinal tract at t_diff = 200ms as opposed to t_diff = 50 ms (p < 0.05). Comparison of DE between DSE and STEAM200 showed a decrease of 12 ± 2% in callosum and of 15 ± 3% in the cortico-spinal tract in STEAM200 (p < 0.05).

Fig. (3). Geometric measures of diffusion: CL, CP and CS are plotted as three phase plot. (a) Comparison of short and long tdiff by STEAM from the same animal. (b) Comparison of DSE and long tdiff by STEAM from the same animal. Light red dots indicate diffusion in voxels at short tdiff = 50 ms, Green dots represent diffusion in the same voxels at tdiff = 200 ms. At long tdiff, diffusion tends to be significantly (p < 0.05) more linearly anisotropic. The red and green squares represent the mean diffusivity at tdiff = 50 ms and 200 ms respectively.

Tractography outcome for ex vivo and in vivo experiments at short and long t_diff were compared (Fig.4). The fibers traced at long diffusion time were much longer than those traced at short diffusion time. The length of fibers and the % increase in the length of fibers at long t_diff are tabulated in Table3 for the fiber tracts in the internal capsule and corpus callosum. In the ex vivo experiments, the callosal fibers were 14.6 ± 5.8% longer for the long t_diff compared to short t_diff. In the in vivo experiments, the callosal fibers were 10.1 ± 2.9% longer. Similarly, the increase in fiber length in the internal capsule was 24.8 ± 6.9% for the ex vivo experiments and was 7.3 ± 2.8% for the in vivo experiments for the long t_diff compared to short t_diff. In comparison with the conventional sequence, the increase in fiber length in STEAM200 with respect to DSE was 6.2 ± 1.5% in the corpus callosum and about 11.5 ± 2.9% in the internal capsule. Overall, in the acquired brain volume, the improvement in tractography was more significant in the ex vivo studies.

Fig. (4). DTI tractography: For the ex vivo study, fiber tracking using (a) DSE, (b) STEAM48 and (c) STEAM192 is shown in the same sample. White arrows and a yellow bracket mark the regions of improvement (longer fibers) using long tdiff. In vivo (d) short and (e) long tdiff tractography results by STEAM sequence were obtained from the same animal. In vivo (f) DSE and (g) long tdiff by STEAM sequence were obtained from the same animal. Arrows mark the regions of improvement (longer fibers) using long diffusion time. (Ex vivo tractography obtained from experimental data in [11])

Protein cross-linking, dehydration, and tissue degradation in formalin fixation post-mortem could alter MRI tissue properties [30, 31]. Fixation shortens tissue T1 and T2 relaxation. Since the T2 values of the brain tissue, in vivo, are much longer than those measured ex vivo, in vivo scans have an advantage of higher SNR per acquisition. In vivo brain T1 is also comparatively longer, translating into better SNR in the DTI with STEAM acquisitions compared to ex vivo.

Fixation decreases ex vivo ADC. The in vivo brain ADC was ~0.8 x 10^-3 mm²/s and a b value of 1200 mm²/s produced e-fold attenuation which produced optimal diffusion weighting. The fixed brain ADC was 0.16 ± 0.02 x 10^-3 mm²/s and a higher b value of 1700 mm²/s was used. Ideally, the b value for the tissue sample with an ADC of 0.16 ± 0.02 x 10^-3 mm²/s would be ~ 4700mm²/s. The optimal b value could not be achieved in the ex vivo experiments because it was constrained by SNR. Importantly, due to the reduced ADC in fixed tissue samples, the t_diff of 48ms results in a rms displacement of 6 μm, whereas the root-mean-squared (RMS) displacement in vivo is 15 μm. At a t_diff of 196 ms in the ex vivo studies, this RMS displacement is approximately 14 μm and about 30 μm for the in vivo studies. Thus, the sampling of the diffusion space is expected to yield more improvement with increasing t_diff for the fixed brain than for the in vivo brain as observed.

The study by D’Arceuil showed no significant change in FA values between in vivo and fixed samples at 7T [32]. However, fixation decreased FA 28% in WM and 10% in GM in our studies. This discrepancy could be due to different degree of tissue degeneration after post mortem [29,32]. Similar to the observation in the aforementioned study [32], the fiber bundles in our study were sparse in the fixed brain tissue compared to in vivo. Fixation also decreased fiber length, consistent with that reported previously.

The geometric measures of diffusion anisotropy serve as excellent measures to determine whether the diffusion tensor in a voxel is linear, planar or spherical. Linear diffusion represents the presence of uni-directional nerve fibers, while planar diffusion represents the presence of multiple fiber directions. Spherical diffusion is equivalent to isotropic diffusion. This study showed that the diffusion along a fiber tract became more linear at long t_diff (200 ms), especially along the fiber terminations of the callosal tract in the cortical surface. Additionally, neighboring voxels showed more inter-voxel coherence of the tensor directions at long t_diff, consistent with the white matter anatomy in the cortical regions.

Tractography in the acquired brain volume revealed that the length of the tracked fibers increased significantly with t_diff = 200 ms when compared with tractography outcome at t_diff = 50 ms. The major nerve fiber pathways such as the corpus callosum and the cortico-spinal tract were tracked. The fiber branches of the callosal tract were much more abundant at t_diff = 200 ms (Fig.3). Further, the fibers traced in the callosum and the internal capsule were much longer at long t_diff than at the conventional diffusion time. The improvements in tractography and DTI characteristics, although significant, were much less in the in vivo data which could be due to lower spatial resolution and fixation effect in the ex vivo study.