In vivo diffusion spectrum imaging of non-human primate brain: initial experience in transcallosal fiber examination

Abstract

In comparison with conventional diffusion tensor imaging (DTI) technique, diffusion spectrum imaging (DSI) allows for delineating crossing and touching fibers in the brain and has been explored in clinical and preclinical studies. Non-human primates (NHPs) resemble most aspects of human and are widely employed in various neuroscience researches and pharmaceutical development. In the present study, a parallel imaging-based DSI protocol was implemented for in vivo fiber tracking of macaque monkey brains on a 3.0 T clinical scanner. Transcallosal fiber tracts of adult macaque brains were examined with DSI and compared with those from a a conventional DTI protocol. The results demonstrate that DSI can reveal the transcallosal fiber bundles much more extensively than the conventional DTI. The preliminary results suggest that DSI may provide a feasible and robust approach for characterizing the fiber pathways in various disease models of NHPs.

Introduction

Non-human primates (NHPs) show high similarity of anatomical structures, functional connections and organizations to humans and are widely used in various studies in neuroscience and pharmaceutical development (1-4). Previous MRI studies have demonstrated that diffusion tensor imaging (DTI) is a robust approach to detect the microstructural abnormality of brain white matter in various NHP models such as stroke, HIV, development, and aging (5-10).

Fiber tractography, which can be non-invasively derived with DTI, illustrates the white matter fiber bundles in brain or spinal cord, showing good consistency with those observed with injected neural tract tracers (11). Currently, DTI and fiber tractography are popularly used on human and animal researches. However, DTI-based fiber tracking is hindered to characterize crossing or touching fibers within a single image voxel (12). In contrast, diffusion spectrum imaging (DSI) allows sufficient angular separation and anisotropy by measuring its diffusion density spectra estimator or the orientation density function (ODF) based on a densely acquired q-space. Hence, it can be used to examine crossing fiber tracts within a single voxel (13,14). DSI has been explored and applied in human studies in recent years (15-19). Prior results of fixed NHP brain tissues have demonstrated DSI could differentiate the cross fibers effectively (20) and showed good agreement between the DSI-identified long association pathways and those with autoradiographic tract tracing in formalin-fixed fascicularis monkey brains (14), suggesting DSI has substantial advantage over conventional DTI approaches and can be a useful tool for examining complex neuronal fiber pathways (14). Postmortem study usually requires sacrifice of live animals for sample collection and does not provide dynamic information. In contrast, in vivo examination allows for monitoring the longitudinal evolution of interested pathways and is much more clinically relevant. To the best of our knowledge, the application of DSI technique in in vivo NHP brain studies has not been reported.

Corpus callosum is the largest white matter structure in the central nerve system. It acts as a bridge between cerebral hemispheres and provides a communicative connection between homologous cortical areas (21). As the malformation of corpus callosum (such as callosal dysgenesis) is usually associated with functional and behavioral abnormality (22,23), it is of great importance to assess the transcallosal connections between the specific cortical regions in the two hemispheres. Based upon Witelson’s classification, corpus callosum can be segmented into seven sections, in which each segment is associated with specific cortical areas (24).DTI-based tractography has shown the capacity to evaluate the abnormal fiber connections to the dorsal and medial cortex, but not the lateral callosal projections in human and macaque monkeys (3), most likely due to the lateral projections crossing through multiple fiber bundles, such as longitude fasciculus.

Current high field clinical scanners provide convenient access for NHP imaging. However, NHP brain is much smaller than human’s (1,200-1,300 cc in human vs. ~100 cc in macaque), the DSI protocols used in human studies must be optimized in order to obtain optimal images of small subjects which require much higher spatial resolution. High resolution DSI of small subjects is readily achieved in dedicated research scanners but not in general whole body clinical scanners due to the limitation of their gradient system. Parallel imaging technique has demonstrated its effectiveness to improve DTI image quality in human and macaque brain studies with clinical scanners (25-28). In the present study, we aimed to establish a DSI protocol for examining NHP brains using parallel imaging technique on a clinical 3T scanner. The feasibility was assessed and demonstrated by tracking transcallosal fibers of adult rhesus monkey brains.

Methods and materials

Animals

Three adult rhesus monkeys (female, 8-12 years old, 9-12 kg) were used. All procedures were approved and in compliance with the Institutional Animal Care and Use Committees (IACUC) of Emory University and the NIH guide for the care and use of laboratory animals. During MRI scanning, animals were anesthetized with 1.0-1.5% isoflurane mixed with 100% O₂ and immobilized with a custom-made head holder in the MRI scanner. Physiological parameters such as End-tidal CO₂, inhaled CO₂, O₂ saturation, blood pressure, heart rate, respiration rate, and body temperature were monitored continuously and maintained in the normal range (29).

MRI experiments

MRI experiments were performed on a Siemens 3T TIM Trio scanner (Siemens Medical Solutions, Inc., PA, USA). To optimize the DSI sequence, DSI images were acquired with an 8-channel volume coil (INVIVO Inc, FL, USA) and 2D stimulated single-shot EPI sequence with the following acquisition parameters: TR =2,500 ms, FOV =96 mm × 96 mm, data matrix =80×80, band width =753 Hz, 1 average, slice thickness =1.2 mm, 30 slices, b =0, 4,000 s/mm². The generalized autocalibrating partially parallel acquisition (GRAPPA) technique was used for image reconstruction (30). Different GRAPPA acceleration factors (R =2, 3, 4) were evaluated for optimization purpose with corresponding minimal TEs =107, 92, 83 ms, respectively. In addition, DSI images without parallel imaging was acquired with the minimal TE =160 ms for comparison purpose.

Optimized DSI protocol was employed to examine the macaque brains with the following imaging parameters: TE =107 ms, TR =7,600 ms, FOV =96 mm × 96 mm, band-width =753 Hz, data matrix =80×80, slice thickness =1.2 mm, 256 diffusion directions covering full q-space 3D grid with 4 radical grid size and b value =4,000 s/mm² and b0 images acquired with b value =0 s/mm², 6 averages, total acquisition time =3 hours. DTI images were acquired with the same setting and a dual spin-echo 2D EPI sequence with GRAPPA (R =3) and the following imaging parameters: TE =109 ms, TR =5,500 ms, FOV =96 mm × 96 mm, band-width =1,350 Hz, data matrix =80×80, 6 averages, slice thickness 1.2 mm, 60 gradient directions with b values of 0 and 1,000 s/mm², total acquisition time =36 minutes. Whole brain field maps were acquired using a gradient echo sequence with TE =6.24 and 8.7 ms, TR =500 ms, FOV =96 mm × 96 mm, data matrix =74×74, voxel size =1.3 mm × 1.3 mm, and slice thickness =1.3 mm. T₁-weighted images were acquired using 3D MPRage sequence with GRAPPA (R =2) and the following parameters: inversion time =950 ms, TE/TR =3.5 ms/3,000 ms, voxel size =1.0 mm × 1.0 mm × 1.0 mm.

Data processing

EPI image distortion correction was performed with the acquired field maps (31), and processed off-line with the FSL package (FMRIB, Oxford, UK) (32).

In order to compare the signal to noise ratio (SNR) of DSI images, the image noise level was defined as the standard derivation (SD) of the background signal outside the brain (33). One representative slice of DSI b0 images was selected for SNR comparison. DSI image SNR was calculated with the mean value of the brain tissue signal divided by the image noise level. The DSI b0 images with different GRAPPA acceleration factors (R =2, 3, 4) and without parallel acquisition were evaluated.

For both DTI and DSI data sets, the white matter tracts were reconstructed using deterministic fiber tracking with 0.5 mm step size (20,34) and the DSI-Studio software (http://dsi-studio.labsolver.org).

Corpus callosum was segmented into seven areas (Figure 1), including rostrum, genu, rostral body, anterior midbody, posterior midbody, isthmus and splenium (24), were registered (with 12 DOF linear affine transformation) to the DSI and DTI b0 images. A macaque monkey template with isotropic 0.5 mm spatial resolution built from averaged T₁-weighed images of adult macaque monkeys in the Yerkes Research Center was utilized. With the seven segments as seeds, transcallosal fiber tracking of DTI and DSI data was performed separately with the DSI-Studio software. The acquired T₁-weighed images were registered (with 12 DOF linear affine transformation) to the DSI and DTI b0 images and used to overlap the DSI and DTI reconstructed fibers.

Figure 1.

Topography of midsagittal corpus callosum. Top, T₁-weighted rhesus monkey brain template with the corpus callosum marked by the red-line; bottom: segmented corpus callosum based upon Witelson’s classification (24). Number 1 to 7: rostrum, genu, rostral body, anterior midbody, posterior midbody, isthmus, and splenium.

Results and discussion

In the present study, a parallel imaging-based DSI protocol was optimized for in vivo rhesus monkey brain study on a 3T clinic scanner equipped with an 8-channel phased-array volume coil and evaluated in transcallosal fiber examination.

Because DSI usually requires stronger diffusion gradient than DTI (currently b=4,000 s/mm² in DSI vs. b=1,000 s/mm² in DTI), the minimal TE can be increased dramatically in clinical scanners which have much lower gradient strength than dedicated animal scanners, resulting in severe susceptibility artifacts. As seen in Figure 2, the usage of paralleling imaging technique significantly reduced the minimal TE and improved the image quality. Also, the highest image SNR (in arbitrary units) was achieved with least parallel acquisition acceleration (SNR =9.3 at R =2), while the SNR was 8.5 at R =3, 7.5 at R =4, and 5.7 without parallel imaging acquisition. In comparison with the whole brain SNR, the SNR in corpus callosum was slightly lower (SNR =8.1, 7.2, 6.0, 4.5 with R =2, 3, 4 and without parallel acquisition, respectively).

Figure 2.

Axial DSI b0 images of an adult rhesus monkey brain with different parallel imaging (GRAPPA) acceleration factors [(A) R =2; (B) R =3; (C) R =4; (D) without parallel imaging].

As illustrated in Figure 3, DSI-derived ODF clearly shows multiple fiber orientations in the interested region of white matter bundles, including dorsal and lateral fibers in the conjunction of primary motor and somato-sensory cortices. In contrast, DTI-derived ODF only exhibits one principle fiber orientation, demonstrating that DSI has the unique capacity for imaging crossing fibers in brain tissue.

Figure 3.

Comparison of DTI and DSI tractography in the white matter region with crossing fibers in an adult rhesus monkey brain. (A) Interested region marked on a macaque brain template; (B) ODF with DTI; (C) ODF with DSI. Abbreaviations: ODF, orientation distribution function; DTI, diffusion tensor imaging; DSI, diffusion spectrum imaging.

The fiber bundles passing through different segments of corpus callosum were examined with DSI and DTI. As demonstrated in Figure 4, the DTI and DSI-derived fiber bundles across rostrum, genu, rostral body, and anterior midbody of corpus callosum are revealed. Obviously, DSI-based tractography revealed more extensive fibers than those from DTI in these interested regions. In addition, long association pathways such as arcuate fasciculus/superior longitudinal fasciculus crossing transcallosal fibers from rostrum, genu and rostral body, were delineated explicitly by DSI (Figure 4A,C,E) but not by DTI.

Figure 4.

Illustration of fiber tractography of an adult rhesus monkey brain using DSI (left) and DTI (right). From top to bottom: DTI- and DSI-derived fibers across rostrum (A,B), genu (C,D), rostral body (E,F), anterior midbody (G,H), respectively. Arrows indicate the color-coding of spatial orientations. DSI, diffusion spectrum imaging; DTI, diffusion tensor imaging.

The fiber bundles across posterior midbody, isthmus, and splenium were evaluated with DSI and DTI as well. Overall, the DSI and DTI tractography results look similar in these segments (Figure 5). However, the DSI tractography from posterior midbody showed longer corticofugal fibers to the brainstem (Figure 5A), and as similar as that from anterior midbody (Figure 4G). In addition, DSI tractography showed more extensive lateral cortical fibers than DTI tractography which only revealed the tracking fibers passing through primary motor cortex (Figures 5A,B). Such differences between DSI and DTI tractography results were also observed in other fiber bundles (Figure 5C-F).

Figure 5.

Illustration of fiber tractography of an adult rhesus monkey brain using DSI (left) and DTI (right). From top to bottom: DTI- and DSI-derived fibers across posterior midbody (A,B), isthmus (C,D) and splenium (E,F), respectively. Arrows indicate the color-coding of spatial orientations. DSI, diffusion spectrum imaging; DTI, diffusion tensor imaging.

Previous results have demonstrated that DTI could partially reveal the transcallosal fiber networks in the macaque brain (3). Although the corpus callosum segmentation in the prior study was slightly different from Witelson’s (24), the transcallosal fibers across the anterior part of corpus callosum mainly comprised the cortices in dorsal orientation (3), consistent with the DTI findings in the present study. In addition, as indicated in the present study, DTI tractography failed to reveal the long association pathways (such as arcuate fasciculus/superior longitudinal fasciculus and descending corticofugal fibers).

Even though NHPs are allowed to be scanned for hours under anesthesia, the current scanning duration may still be a concern in some disease models and hence hinder the DSI’s application in translational researches. As raw image SNR is the major consideration in DSI data acquisition, a low GRAPPA acceleration factor (R =2) and low bandwidth were selected and applied in the present study. To further reduce the scanning duration but with the image SNR maintained, a better phased-array coil (with more channels and/or optimal dimensions) can be used in order to reduce the number of signal averages. Also, DSI data acquisition may be accelerated by using compressed sensing (35) or reduced-encoding DSI (36).

In addition, the DSI image quality may be further improved by using multi-shot EPI acquisition (37), spiral acquisition (38), iterative self-consistent parallel imaging reconstruction strategy (39), or steady-state radial acquisition (40).

In conclusion, a parallel imaging based DSI protocol for in vivo macaque brain imaging was implemented on a clinical setting. The preliminary results demonstrate its capacity to examine the white matter fiber bundles with crossing fibers and visualize long fiber pathways and a comprehensive map of neuronal circuits. The novel approach may allow for evaluating the integrity of fiber pathways and understanding the neural substrates in NHP models of various diseases and neuroscience researches.