Non-normalized individual analysis of statistical parametric mapping for clinical fMRI
Correspondence Address: Source of Support: None, Conflict of Interest: None DOI: 10.4103/0028-3886.82714
Source of Support: None, Conflict of Interest: None
Background : Pre-operative evaluation to localize function within the cerebral cortices is essential before brain surgery. Blood oxygenation level-dependent functional magnetic resonance imaging (fMRI) has been used for this purpose. Aims : To obtain clearer and more understandable functional images. Patients and Methods : Ten patients with brain tumors underwent fMRI including hand-gripping and word generation tasks. The statistical parametric mapping (SPM) approach was used for subsequent analysis to localize the motor or language functions. SPM includes image pre-processing, statistical computation, and significance testing. In order to demonstrate a spatial relationship between the lesions and a functioning area in the individual structural MR images, normalization to the Montreal Neurological Institute coordinates was intentionally not performed. Results : In seven cases out of 10, the patient's motor area was clearly visualized. Language areas were also demonstrated in seven cases. Conclusions : We conclude that application of SPM (version 8) analysis to non-normalized individual data for the purpose of performing pre-operative fMRI is a useful method for investigation of functional localization.
Keywords: Analysis, clinical, functional magnetic resonance imaging, non-normalized, statistical parametric mapping
Blood oxygenation level-dependent functional magnetic resonance imaging (BOLD fMRI) is an established method for visualizing brain activity. In fMRI, the neuronal activity is visualized as an increase in cerebral blood flow (CBF).  This signal change is weak enough to be influenced by head movement. To obtain more accurate data, correction of this movement is necessary. The important principle of the fMRI is that it measures changes in CBF, and is therefore an indirect measure of the neuronal activity. Hence, the regions demonstrated by fMRI are possible areas activated by the stimulation and a statistical approach is essential. The statistical parametric mapping (SPM; Wellcome Trust Centre for Neuroimaging) developed by Friston is widely used.  Results obtained from SPM provide clear and accurate information.
From 2009 to 2010, a total of 12 brain tumor patients underwent fMRI while performing both motor and verbal tasks. Of the 12 patients, one patient with severe myopia and another patient who could not read a letter were excluded. Thus, 10 patients, seven men and three women, were analyzed in this study. The mean age was 50.6 ± 14.6 years. There were nine patients with intramedullary tumor and one patient with meningioma [Table 1].
MR images were obtained using a Achieva 3T system (Philips, DA Best The Netherlands). BOLD functional images were acquired using single-shot gradient-echo echo-planar imaging (EPI) (repetition time = 3000 ms, echo time = 35 ms, flip angle = 90 degrees, slice gap = 0 mm, field of view = 200 mm, voxel size = 2.5 mm × 2.5 mm × 4 mm). For structural images, we used thin-sliced gadolinium-enhanced T1-weighted images.
For investigations using fMRI, a blocked design and event-related design are available.  All actions to be performed during imaging were instructed by visual cues.
In this design, the trials from each condition are grouped together in time to form blocks.  In our blocked motor paradigm, the patients perform a gripping task (grasping their hand repeatedly every 5 s) for 30 s after an initial resting time of 30 s. These 30-s periods of gripping or resting are the blocks. In this design, we analyze the CBF change between the resting blocks and the gripping blocks.
The concept of the event-related design is that the processes of interest can be evoked transiently by brief presentations of individual stimuli.  In our motor paradigm, patients will grip at 30, 35, 40, 45, 50, and 55 s from the beginning of the scan in the first working block, 90, 95, 100, 105, 110, and 115 s in the second block, and 150, 155, 160, 165, 170, and 175 s in the last block.
Patients grasped a sponge ball with their hand contralateral to their brain lesion. During 180 s of scanning, the patients grasped a ball at 30, 35, 40, 45, 50, 55, 90, 95, 100, 105, 110, 115, 150, 155, 160, 165, 170, and 175 s from the beginning according to a visual cue in the center of the screen. Then, the data were analyzed in two ways.
We used a word generation task to assess the language function. The patients were instructed to silently generate a word beginning with a single presented Japanese hiragana character. Letters were projected one by one in a random order at the center of the screen at 30, 34, 38, 42, 46, 50, 54, 58, 60, 90, 94, 98, 102, 106, 110, 114, 118, 120, 150, 154, 158, 162, 166, 170, 174, and 178 s. Each letter was presented on the screen for 1 s.
Re-alignment and reslicing the functional images
The role of this procedure is to remove movement artifacts in the functional images. Sixty images of the patients were taken in each slice. In order to reduce motion artifacts, re-alignment to the first image was performed according to six parameters, including x, y, z-translations and x, y, z-rotation.
Segmentation of structural images
The structural images were segmented into different tissue classes containing grey matter, white matter, and cerebrospinal fluid in order to correct bias.
Co-registration of adjusted functional and structural images
Structural and fMR images were taken for each participant. The activating area should be indicated on the structural images because echo planar functional images have a geometric distortion and low resolution. As tumor resection will be performed according to the higher resolution structural images, co-registration of functional imaging into structural images is necessary.
Smoothing of the co-registered images
Spatial smoothing was performed to increase the signal to noise ratio (S/N ratio). The full-width at half maximum (FWHM) was set for 8 mm in all three directions (x, y, and z).
We performed the first-level single-subject analysis based on the general linear model (GLM) implemented in SPM8. We used t-maps from the event-related design generated from the tasks. For detection of the motor area, the statistical maps were thresholded at P<0.001 without correction for multiple comparisons. If activation was undetectable, the P-value was changed to <0.05 and the analysis was repeated. For detection of the language area, the statistical maps were thresholded at P<0.05 using a family-wise error (FWE) correction for multiple comparisons. When the activated area was undetectable, the map was thresholded again at P<0.001 or 0.01 without correction for multiple comparisons.
In seven patients, the contralateral hand motor cortex was detectable in the pre-frontal gyrus [Table 1]. In six patients, the hand area was detected with a P-value no greater than 0.001. In case 1, the P-value was raised to <0.05 to detect the area. In cases 4, 6 and 10, we were unable to identify the hand area. Of the three patients in whom the hand representation was undetectable, cases 4 and 6 demonstrated hemiparesis. Case 10 could not grasp a ball because she did not understand the task.
In nine of the 10 patients, the language area was detected in the inferior frontal gyrus [Table 1]. Initially, we used a t-test with FWE corrected (P<0.05). The language area was detected in four patients (cases 1, 5, 6, and 10). We could then detect the language area in three additional patients (cases 3, 7, and 8) at P<0.001 without correction for multiple tests and in two patients (Cases 4 and 9) this was uncorrected (P<0.01), as shown in [Table 1]. The language area was located in the left frontal lobe in eight patients and bilaterally in one patient with left amygdala dysembryoplastic neuroepithelial tumor (Case 1).
Case 7: Left frontal lobe astrocytoma
A 26-year-old man with intractable complex partial seizure was diagnosed with recurrent left frontal astrocytoma. The patient underwent an fMRI study before re-operation for lesion removal. The hand area was detected with t-maps (P<0.001, uncorrected) and the language area was detected in the left frontal lobe with t-maps (P<0.001, uncorrected) as shown in [Figure 1]. One week after the lesionectomy, he had remission of the seizures.
MRI using BOLD contrast  was developed in 1990 by Ogawa et al. and has been widely used for brain function visualization. The SPM analysis involves pre-processing of the images, statistical computation, and statistical testing. Pre-processing should be performed before data analysis to obtain more accurate results.
In the field of neuropsychology, many task designs such as subtraction designs,  conjunction designs, , parametric designs, ,,, and factorial designs , are used. However, for neurosurgical applications, simple tasks for detection of the motor cortex and language are sufficient. Thus, we have adopted a simple subtraction method. In the subtraction design, we detect CBF differences between the resting and the behaving states. Patients cannot endure long periods of testing. In order to make the examination shorter and more efficient, we designed a grasping task and analyzed results as both blocked and event-related designs.
We were able to detect the hand area in seven patients. The three patients in whom the hand area was not detectable resulted from motor weakness in those patients or poor task comprehension. The location of the hand motor area in the pre-central gyrus is well known, and variation from this position is extremely rare. Thus, grip tasking can be used reliably regardless of whether patients can perform the tasks appropriately.
Detection of the language area is more difficult than localizing the motor area. In order to detect the language area, we initially chose FWE correction; however, this option is relatively strict for area detection in some cases. Hence, when the language area could not be localized in this way, the next step was to loosen the statistical criteria to P<0.001 without correction, similar to the motor analysis. Considering its invasiveness, fMRI has been applied for clinical use. Kamada et al. performed both the Wada test and fMRI, and reported that the task accomplishment rate was 84.6% and the rate of successful detection was 90.1% by fMRI.  This suggested that language lateralization could be determined in 76.2% of the patients using fMRI.
We have used two major experimental designs, blocked and event-related designs.We performed both blocked design and event-related analyses. In the blocked design, a condition is provided continuously for an extended time interval to maintain cognitive engagement, and different task conditions are usually alternating in time.  This design is simple but very powerful and has the advantage of intense, , relatively large BOLD signal change relative to baseline  and increased statistical power.  With a long block length of more than 10 s, the blocked design analyses have the advantage of detecting statistically significant brain activities regardless of the timing or shape of the BOLD signals.  Although the blocked design is very useful, it cannot estimate the time course of activation in active voxels.  Event-related designs can provide more detailed information about the timing and shape of the hemodynamic response. In this design, discrete and short-duration events are indicated with timing and order.  It can detect transient variations in hemodynamic responses, allowing for analysis of individual responses to trials,  reducing a subject's expectation effects  and is less sensitive to head movement.  Despite these advantages, the event-related design has an inferior detection power.  It was reported that event-related designs have the potential to provide comparable or even higher detection power over the blocked designs for localizing language function in brain tumor patients. 
Selection of statistical analysis options remains controversial. The SPM default statistical analysis settings have a threshold of P<0.05 with FWE correction and P<0.001 without correction. Initially, we used these default options, and reduced the threshold to re-analyze when a statistically significant area was not identified. This procedure will be necessary because the aim of pre-operative fMRI should be preservation of function. Therefore, underestimation of the size of a functioning area should be avoided. An advantage of SPM is that group analysis can be performed. In order to perform group analysis, each subject's data should be into Montreal Neurological Institute (MNI) coordinates.Brain anatomical structure will be intact, the coordinate gap between the original brain and the MNI will not be so disparate. On the other hand, normal brain tissue will be compressed and displaced in the patients with brain tumor. Therefore, conversion into MNI coordinates is not suitable because it will result in even larger differences. Moreover, pre-operative fMRI is required to get information about the relationship between the lesion and the functional region. Therefore, the procedure of normalization is not appropriate for pre-operative evaluation.
There exist alternative methods of assessing fMRI data, such as FSL and Brain Voyager.  The Analysis Group at the Oxford Centre for Functional MRI of the Brain (FMRIB) developed FSL written in C.  Despite the existence of these tools, SPM has been widely used because of its convenience and long history.
This study suggests that non-normalized individual SPM analyses provide clear information and have the potential to improve the success rates of function-preserving operations.