NHP Issues
Even though the main gist of this guide concerns data processing, it is worthwhile to understand some of the pre-data processing issues related to NHP neuroimaging as these will give valuable insights into the source of the data processing issues that are typically encountered.
Several lengthy steps are necessary before being able to put a NHP individual in an MRI scanner. Here is a brief summary of those. An important implication is that typically NHP studies can recruit a lower number of subjects than in human studies. A corollary to this point is why it is so crucial that initiatives such as PRIME-DE exist to aggregate data from different centers to leverage the challenge of accessing more NHP data.
-
Study preparation: Ethical approval (3R), “Recruitment” of non-human primates (find provider, transportation procedure, temporary house next to MR facility if the animal provider is far away, rearing and training for awake experiments etc.).
-
Animal handling and preparation: Animal transportation to the MR room, anesthesia before and during the acquisition if needed, surgery for headpost fixation for awake task fMRI acquisition. A vet is also required to be present for the handling and the medical support during the acquisition. As a result, a huge amount of out-of-scanner preparation and time and cost is involved for each animal scan session.
-
MR coil decision: Whether to build a home-made coil, buy a non-human primate coil already designed by a company or another lab, or to use clinical coils designed for humans (See Figure 1 below for examples). This is in contrast with human MRI scans where a highly standardized birdcage coil is typically used and broadly available. Unfortunately, due to different head size and shape (one obvious difference between human and NHPs such as macaque or baboon is the elongated of ape’s mouth), those human coils are far from optimal from NHP studies and may even not fit one ape head even if the ape brain is much smaller than the human brain! As a result, NHP MRI data often shows huge intensity inhomogeneity due to the inadequacy of the head size and the coil elements positioning (Figure 2). Surface coil may also be chosen for NHP MRIs, especially in the case of scanning using headpost or stereotaxic frames. The advantage of surface coils is the very strong receive signal in the vicinity of the coil but at the expense of a high intensity bias field. This strong intensity bias field, if not well taken care of (see part C 5.), will be problematic for processing stages such as skull stripping, registration and normalization. Also depending on the acquisition set-up, the head of the NHP individual may be far from the magnet isocenter, leading to image distortions due to the gradient non linearity in such areas.
|
|---|
| Figure 1: Different examples of acquisition setups with different coils: A: Macaque 24-channel receive coil for 3T clinical scanner (Autio et al. NeuroImage (2020)), B: Human knee coil used for animal imaging at 3T (here with a rabbit brain, works for macaque too!) (Park et al. J Korean Soc Radiol (2017)), C: 8Tx/24Rx monkey coil for 7T MRI (Gilbert et al. NeuroImage (2016)), D: Embedded 8-channel coil for 3T imaging in a clinical scanner (Janssens et al. NeuroImage (2012)), E: Quadrature volume coil for monkey in sphinx position at 3T (Roopnariane et al. Concepts in Magn Reson part B (2012)), F: Monkey sitting in sphinx position for an awake fMRI experiment using an 8 channel phased array coil (Cotterau et al. Cereb. Cortex (2017)) |
|
|---|
| Figure 2: Example of strong intensity bias due to the use of surface coils for acquisition. Note how the thick temporal muscles “blocks” signal to the receive coil from the brain, diminishing SNR towards the center of the brain |
-
Non-standard orientations: Depending on acquisition settings (anesthesia, coil choice), NHPs can be scanned in the typical human position (head first, supine) or in non-standard position, for instance in the sphinx position (in their MRI chairs or in stereotaxic frames). MR vendors typically do not expect this position for the patient being scanned and as a result, during the patient registration on the MR console, the researcher must choose an incorrect position for the NHP being scanned (for instance “head first - supine” while the NHP is in actually in sphinx position) and this will result in incorrect labels for the acquired image (see example in figure below in part C 1.).
-
Anesthesia: The choice of anesthesia will influence the length of the acquisition as well as the BOLD signal itself. Physiological monitoring is required throughout the acquisition and physiological recording may be used for subsequent data processing steps.
-
Contrast agent: NHP fMRI studies often rely on injection of a iron-based contrast agents such as monocrystalline iron oxide nanoparticle (MION), which strengthen T1 and T2/T2*, the latter being used to increase the sensibility for detecting brain activation. In addition, shorter T2* shortens the optimal TE for brain activation related signal changes, permitting higher temporal resolution. Some specific issues are related to the accumulation of the MION agent in animals after repetitive injections across sessions: MION blood-level needs to be assessed at the beginning of each session (through adaptation of contrast agent dosis and measurement of SNR and T2* during each session), and chelating the residual MION in the animals after acquisition campaigns. Useful References: Leite et al. NeuroImage (2002), Vanduffel W, et al. Neuron (2001), Mandeville et al. Magn Reson Med (1998). Review: Wu et al. NMR Biomed (2004).
-
Variable Field-of-View (FOV): Animal sizes vary greatly across species (e.g. macaque, marmoset etc.) but also within the same species (e.g. male vs female, older vs younger). See for instance the size of the temporal muscles in older males compared to young females (Figure 3). Depending on the size heterogeneity within the population being scanned, acquisition parameters must be optimized to adjust the field-of-view to each animal caliber and therefore different parameter settings may exist for a single study. This parameter heterogeneity may differ even more so between sites. In addition, due to the high size of the matrix, for very high resolution acquisitions (0.4 mm isotropic for instance), the FOV is often very tight around the head of the animal and a folding of the image is almost inevitable. One strategy is to tilt the FOV so that the folding only compromises the non-brain areas. The resulting images will be oblique, and additional pre-processing steps will need to be implemented (see details on how to handle oblique acquisitions in section C 1).
|
|---|
| Figure 3: Anatomical differences between female and male macaques. Left: T1w image of a female macaque. Right: T1w image of a male macaque with much bigger temporal muscles. Of note the male specimen on the right was involved in an awake fMRI experiment and was implanted with a headpost before this MRI scan. |
-
SNR: Due to the smaller head size compared to humans, higher resolutions are required for NHP imaging, resulting in lower SNR for images. Strategies to compensate for this is to average scans, either within the same acquisition (the averaging is then done online) or by repeating the same scan (thus the averaging has to be done offline). A correlation to that is that NHP imaging is necessitating longer acquisition times and sometimes there is just not enough time to get to SNR similar to human images. Longer acquisition times also come with additional drawbacks: magnetic field drift, B0 inhomogeneity in EPI acquisition due to bigger matrices and longer echo spacings. This B0 inhomogeneity would be even more problematic for acquisition settings where the head of the primate is not at the magnet isocenter, as would be the distortions due to non linearity gradient distortions
-
Subject Motion: Less motion than in human if anesthesia is used, more motion if it is awake scanning of a non-compliant individual despite using headpost. In awake scans, the naturalistic viewing paradigm (e.g. movies, Inscapes, etc.) appears to improve the head motion as compared to resting state (Figure 4).
|
|---|
| Figure 4: Examples of head motion (Framewise displacement) in awake and anesthetized scans in macaque (Top). Compared to resting state, the naturalistic viewing paradigm (e.g. movies, Inscapes) improves the head motion in awake scans (Bottom). |
- NHP BIDS (Brain Imaging Data Structure): Usually DICOM format comes out of clinical scanners. Although BIDS has recently become a global standard for human neuroimaging, the conversion of NHP DICOM data to BIDS is less frequent and can sometimes be problematic if the sequence designation on the scanner has not been designed for this purpose. One of the reasons why BIDS has become so quickly and globally accepted is the motivation for researchers to easily share their data in OpenNeuro and use BIDSApps. This motivation has been less prominent in NHP imaging, although PRIME-DE structures the data in the BIDS format and PRIME-RE is a website with NHP-specific pipeline tools, some of which understands the BIDS data format. However, BIDS may require an additional bifurcation for NHP imaging as less standardized sequences may not be covered by the current BIDS specification and additional metadata fields may be required.