fMRI in Surgical Planning: The talk will focus on contrasting task and resting state fMRI for use in pre-surgical planning. The uploaded file focuses on pre-processing of fMRI data.
Target Audience: This session is intended for Clinicians or Scientists interested in learning the basics and implementing an fMRI program for Pre-surgical planning.
Objectives: Following this session the learner will be able to make informed decisions on which type of fMRI (task vs. resting state) to use for their practice, and what the respective advantages and disadvantages are of the different methods.
2.1 MRI and Clinical Protocols Patients with newly diagnosed brain tumors are scanned using a 3T Scanner. The resting state data are acquired using a T2* echo plane imaging (EPI) sequence (3 × 3 × 3 mm cubic voxels; 128 volumes/run; TE = 27 ms; TR = 2.8 s; field of view = 256 mm; flip angle = 90°), while the patients are instructed to remain still and fixate on a visual cross-hair without falling asleep (2 runs of 6 minute each for a total time of 12 minutes). Tumor protocol anatomic imaging also includes T1-weighted magnetization prepared rapid acquisition gradient echo (MP-PAGE), T2-weighted fast spin echo, susceptibility-weighted imaging (SWI), diffusion-weighted imaging (DWI) and pre and post gadolinium contrast T1-weigted fast spin echo in multiple projections. All anatomic and functional magnetic resonance data are acquired in less than 60 minutes. The processing of the rsfMRI data is performed using a locally developed set of algorithms [13] with details provided in http://4dfp.readthedocs.io. This processing, with its respective quality control (QC) measures, is implemented within the Translational Imaging Platform (TIP), a custom XNAT-based informatics system [14]. The TIP includes an interface to query for and retrieve patient studies from the clinical Picture Archiving and Communication System (PACS) and pipelines to fully automate the rsfMRI processing. The pipelines generate DICOM-formatted images of the resting state networks and a web-based QC report. Trained technicians manage the overall workflow and for studies that pass QC review, the generated DICOM images are pushed back to the PACS for use by clinicians. Studies that fail QC are reprocessed with adjusted parameters. In rare cases, the study is unable to be processed, typically due to severe subject motion, image artifacts, or poor spatial alignment with anatomic scans. Once the generated images are available on the clinical PACS, a neuroradiology fellow integrates them with the task fMRI, tractography, and anatomic images on a surgical planning station. The entire examination is reviewed by a neuroradiology attending, a report is dictated, and the intraoperative navigation-compatible fused imaging is transmitted to the operating room. If the quality of the rsfMRI data is determined to be sub-optimal (either by the automatic QC measures or during the review by the neuroradiology fellow and attending) this will be included in the written report and the neurosurgery service will be notified. This set of procedures was refined over time in response to feedback from the personnel that implement the process and from the neurosurgeons that are the end users of this information. The service is frequently used by the neurosurgeons that perform tumor surgery at our institution. The rsfMRI information is especially valuable if the task fMRI is of poor quality or is non-existent, such as can occur when patients are not able to cooperate with the task fMRI requirements.
2.2 Overview of Processing Methods for rsfMRI Investigators interested in pursuing rsfMRI experiments need to be aware of certain practical considerations. A new investigator may wish to use one of the freely available software packages to help process their data. The following three packages include all the basic operations needed to analyze rsfMRI data: FSL [15], AFNI [16], SPM www.fil.ion.ucl.ac.uk/spm/. Each package is downloadable at no cost, is documented in an easily accessible web site, provides on-line support, and is associated with regularly scheduled training workshops. FSL and AFNI are the most modular. Modular organization requires the user to understand the sequence of analysis operations and discourages "push-button" or rote analyses. Basic preprocessing common to both task-based and resting state fMRI must precede all subsequent analyses (detailed below). Once basic fMRI preprocessing is complete, the investigator must choose between two principal modes of rsfMRI analysis: (i) seed-based correlation (SBC) mapping and (ii) spatial independent component analysis (sICA). SBC mapping is well suited to investigations of the functional connectivity of a priori targeted regions of interest but requires extensive preprocessing to reduce the impact of artifact (see below). sICA provides a direct means of separating artifact from neural signals [17] but is less well suited to investigating targeted regions of interest. Both SBC and sICA yield highly reproducible results at the group level [18]. RSNs near the top of the hierarchy, especially, the DMN, appear very much the same whether mapped by SBC or sICA; at lower levels of the hierarchy, RSN topography derived by the two methods systematically differs. Some networks are lateralized using sICA analysis, whereas SBC mapping generally yields highly symmetric maps [19, 20]. Modest asymmetry can be observed with SBC if seeds are placed in homotopic parts of the brain instantiating language (left hemisphere) and orienting (external capture of attention; right hemisphere) [19]. This document will focus on a SBC style methodology used by our lab implemented as a series of preprocessing steps followed by classification using a previously trained multi-layer neural network. This method uses prior information for the localization of RSNs across the brain, and requires careful de-noising to exclude non-neural artifact from the data. An advantage of SBC in comparison to ICA is that using prior information provides a more robust mapping in individual patients with relatively high motion and relatively short scanning times that are more realistic in a population of patients with brain tumors [21].
2.3 General Preprocessing The following preprocessing steps are performed on the raw fMRI data [22]: 1. Compensation for slice dependent time shifts: Although a typical TR is around 2s (faster for the newer multi band sequences), each slice of the brain is acquired at a slightly different time and thus interpolation methods are used to create a full brain image that was acquired “simultaneously” at a single time point. 2. Elimination of systemic odd-even slice intensity differences due to interleaved acquisition. 3. Rigid body correction for head movement within and across runs using affine registration techniques. 4. Intensity scaling (multiplicative factor applied to all voxels of all frames within each run) to obtain a mode value of 1000 [23]. This facilitates assessment of voxel-wise variance for the purpose of quality assessment but does not affect computed correlation. 5. Head movement correction is achieved by atlas transformation using a composition of affine transforms connecting the fMRI volumes with the T1- and T2-weighted structural images. Head movement correction is included in a single resampling to generate a volumetric time series in 3 mm cubic atlas space. 6. Spatial smoothing using a 6 mm full width half maximum Gaussian blur in each direction in order to increase the signal to noise with some loss of resolution. 7. Voxel-wise removal of linear trends over each run. 8. Temporal band pass filtering to retain frequencies between 0.01 and 0.1 Hz. 9. Reduction of spurious variance by regression of nuisance variables. These include waveforms derived from: a. Signal of head motion correction. b. Signal extracted from CSF. c. Signal extracted from white matter areas. d. Global signal across the whole brain. The last step of global signal regression (GSR) remains controversial and further detail is provided in the next section. 10. Removal/Scrubbing of high motion time points to minimize the impact of head motion on the correlation results. Two common strategies to identify compromised frames include selection based on analysis of retrospective head motion correction timeseries [24] or by voxel-wise evaluation over the whole brain of the differentiated fMRI timeseries (DVARS measure) [25, 26].
2.4. Global Signal Regression Global signal regression (GSR) prior to correlation mapping is a highly effective means of reducing widely shared variance and thereby improving the spatial specificity of computed maps [27-29]. Some part of the global signal is of neural origin [30]. However, much of the global signal represents non-neural artifact attributable to physical effects of head motion [31-34] and variations in the partial pressure of arterial carbon dioxide [35]. Absent GSR, all parts of the brain appear to be strongly positively correlated [36-39]. GSR causes all subsequently computed correlation maps to be approximately zero-centered; in other words, positive and negative values are approximately balanced over the whole brain [27]. Thus, GSR does negatively bias all computed correlations, although iso-correlation contours, i.e., map topographies, remain unchanged. This negative bias has caused some to criticize GSR on the grounds that it induces artifactual anti-correlations [40, 41]. Although it has since been demonstrated that some parts of the brain appear to be truly anticorrelated in the resting state, as demonstrated using sICA [42, 43]. More recent objections to GSR focus on the possibility that it can distort quantitative functional connectivity differences across diagnostic groups [44]. However, this objection to GSR is irrelevant in the context of using rsfMRI for purposes of RSN mapping in individuals.
1. Gulati, S., et al., The risk of getting worse: surgically acquired deficits, perioperative complications, and functional outcomes after primary resection of glioblastoma. World Neurosurg, 2011. 76(6): p. 572-9.
2. Lacroix, M., et al., A multivariate analysis of 416 patients with glioblastoma multiforme: prognosis, extent of resection, and survival. J Neurosurg, 2001. 95(2): p. 190-8.
3. McGirt, M.J., et al., Independent association of extent of resection with survival in patients with malignant brain astrocytoma. J Neurosurg, 2009. 110(1): p. 156-62.
4. Dandy, W.E., The treatment of brain tumors. JAMA, 1921. 77: p. 1853-1859.
5. Ojemann, G.A., Functional mapping of cortical language areas in adults. Intraoperative approaches. Adv Neurol, 1993. 63: p. 155-63.
6. Petrella, J.R., et al., Preoperative functional MR imaging localization of language and motor areas: effect on therapeutic decision making in patients with potentially resectable brain tumors. Radiology, 2006. 240(3): p. 793-802.
7. Spitzer, M., et al., Category-specific brain activation in fMRI during picture naming. Neuroreport, 1995. 6(16): p. 2109-12.
8. Biswal, B., et al., Functional connectivity in the motor cortex of resting human brain using echo-planar MRI. Magn Reson Med, 1995. 34(4): p. 537-41.
9.Kokkonen, S.M., et al., Preoperative localization of the sensorimotor area using independent component analysis of resting-state fMRI. Magn Reson Imaging, 2009. 27(6): p. 733-40.
10. Liu, H., et al., Task-free presurgical mapping using functional magnetic resonance imaging intrinsic activity. J Neurosurg, 2009. 111(4): p. 746-54.
11. Shimony, J.S., et al., Resting-state spontaneous fluctuations in brain activity: a new paradigm for presurgical planning using fMRI. Acad Radiol, 2009. 16(5): p. 578-83.
12. Smith, S.M., et al., Correspondence of the brain's functional architecture during activation and rest. Proc Natl Acad Sci U S A, 2009. 106(31): p. 13040-5.
13. Smyser, C.D., et al., Longitudinal analysis of neural network development in preterm infants. Cereb Cortex, 2010. 20(12): p. 2852-62.
14. Marcus, D.S., et al., The Extensible Neuroimaging Archive Toolkit: an informatics platform for managing, exploring, and sharing neuroimaging data. Neuroinformatics, 2007. 5(1): p. 11-34.
15. Jenkinson, M., et al., Fsl. Neuroimage, 2012. 62(2): p. 782-90.
16. Cox, R.W., AFNI: what a long strange trip it's been. Neuroimage, 2012. 62(2): p. 743-7.
17. McKeown, M.J., L.K. Hansen, and T.J. Sejnowsk, Independent component analysis of functional MRI: what is signal and what is noise? Current opinion in neurobiology, 2003. 13(5): p. 620-9.
18. Damoiseaux, J.S., et al., Consistent resting-state networks across healthy subjects. Proc Natl Acad Sci U S A, 2006. 103(37): p. 13848-53.
19. Fox, M.D., et al., Spontaneous neuronal activity distinguishes human dorsal and ventral attention systems. Proc Natl Acad Sci U S A, 2006. 103(26): p. 10046-51.
20. Salvador, R., et al., Neurophysiological architecture of functional magnetic resonance images of human brain. Cereb Cortex, 2005. 15(9): p. 1332-42.
21. Lee, M.H., et al., Clinical Resting-state fMRI in the Preoperative Setting: Are We Ready for Prime Time? Top Magn Reson Imaging, 2016. 25(1): p. 11-8.
22. Shulman, G.L., et al., Right hemisphere dominance during spatial selective attention and target detection occurs outside the dorsal frontoparietal network. J Neurosci, 2010. 30(10): p. 3640-51.
23. Ojemann, J.G., et al., Imaging studies of memory and attention. Neurosurg Clin N Am, 1997. 8(3): p. 307-19.
24. Power, J.D., et al., Methods to detect, characterize, and remove motion artifact in resting state fMRI. Neuroimage, 2014. 84: p. 320-41.