Posted: August 1st, 2023

Comparison of Resting-State fMRI Data Preprocessing Techniques in Patient

Comparison of Resting-State fMRI Data Preprocessing Techniques in Patient with Brain Tumor
Name

Course
Tutor
Date

Comparison of Resting-State fMRI Data Preprocessing Techniques in Patient with Brain Tumor
Introduction
Introduction to Tumor Dataset
A consistent, evidence-based data reporting concerning cancer-related diagnosis was developed by the International Collaboration on Cancer Reporting (ICCR) datasets. The datasets produced from the assessment of different types of tumors are reported and documented in a consistent style across all individual tumors using all parameters required for assisting and guiding the management and prognostication of a particular diagnosis. Brain Tumor is considered one of the acute diseases that are associated with a high mortality rate. Acquiring data regarding brain tumors has increasingly become essential in managing and predicting brain tumor diagnosis. Various computer-aided diagnosis systems have been adopted globally for detecting and classifying brain tumors. The systems engage in four primary stages in the detection and classification process. The stages include preprocessing, segmentation, extraction of features, and classification. Preprocessing is an essential stage as it sets the level by which the rest of the stages and quality of data are achieved. Preprocessing involves eliminating the medical resonance images (MRI) noise resulting from light reflections or inaccuracies in the imaging medium. The elimination of the medical resonance images (MRI) is conducted through the Magnetic resonance Imaging (MRI) data processing tool.
Main Preprocessing Methods
Magnetic resonance imaging (MRI) data processing is utilized to eliminate the medical resonance images in brain tumor clinical testing and data processing. MRI is divided into two main preprocessing methods; task-based functional MRI (tb-fMRI) and resting-state functional MRI (rs-fMRI). The task-based functional MRI (tb-fMRI) is widely used in clinical practices and neurosurgical oncology patients with a brain tumor. tb-fMRI provides the clinician with the ability to preoperatively plan and facilitate a safe and maximal surgical resection of tumors in the brain’s eloquent areas (Kumar et al., 2020). Although tb-fMRI is widely used, it has various drawbacks and limitations concerning its mapping. One of the apparent limitations of tb-fMRI is the requirement to combine multiple language paradigms for language mapping (Petrella et al., 2006). The language paradigms are used for generating reliable and accurate activation of language networks that enable tb-fMRI to map language areas in the brain.
Additionally, using language paradigms to achieve tb-fMRI language mapping involves highly trained personnel capable of determining patient cognition and choosing the most suitable language paradigms. Another throwback of tb-fMRI mapping is associated with the involvement of the patient in the experimental and data acquisition process (Sharaev et al., 2018). Tb-fMRI preprocessing depends on a patient’s psychophysiological state and ability to activate a particular brain area for data acquisition, which becomes impossible with a patient under sedation.
The advances that have been achieved in recent years in functional magnetic resonance imaging have seen the development of resting-state functional magnetic resonance imaging (rs-fMRI). rs-fMRI is a robust adjunct to tb-fMRI, providing a more prolific experience and evading the limitations and throwbacks of tb-fMRI mapping. rs-fMRI focuses on spontaneous low-frequency fluctuations in blood oxygenation level-dependent (BOLD) signal (<0.1 Hz). Brain mapping primary target specific areas with synchronous low-frequency fluctuations occurs when cognitive, language and motor tasks are not performed by the brain (Lee et al., 2016). The ability to evaluate low-frequency fluctuations in BOLD enables the successful performance of rs-fMRI when the patient is under anesthesia or sedation. Another factor that provides rs-fMRI data preprocessing methods in a patient with a brain tumor an advantage is the ability to map more general structure of the functional network. Additionally, in rs-fMRI mapping, the RSNs involved in the sensorimotor network and eloquent cortex have been defined, including the visual and auditory networks. rs-fMRI is considered the most preferable method in clinical studies on brain tumors as it offers a powerful tool that can detect different patterns among healthy individuals and those diagnosed with various types of brain tumors. Additionally, changes in the values of rs-fMRI data preprocessing techniques including mAFLL, fALFF, ICA, and ReHo within the brain have enabled the identification of a new number of neuropsychiatric disorders, such as depression, autism, schizopherenia, and attention deficit hyperactivity disorder.
Effect of Preprocessing On Data
The effect of preprocessing on data depends on the preprocessing applied. The application of rs-fMRI data preprocessing techniques in patients with a brain tumor has been identified to have various effects on data. According to Vakamudi et al. (2019), the rs-fMRI data preprocessing techniques have significantly higher accuracy rates than the-MRI and tb-MRI data preprocessing techniques. Studies conducted by Zhang et al. (2009) and Shimony et al. (2009) on patients diagnosed with different neurological diseases reported high data to overlap effect between rs-fMRI and tb-fMRI preprocessing in the comparison of stimulation mapping and motor network. However, according to a study conducted by Tie et al. (2013), the rs-fMRI has a more variable effect on data when mapping language systems. According to Sair et al. (2016), one of the recent works on brain tumors indicated that only rs-fMRI group level at moderate and tf-MRI language network managed to provide a concordance effect on data recording a substantial subjective-level variability. The effect of preprocessing on data has continued to increase the availability and evolution of high-speed data acquisition methods that have significantly reduced sensitivity to physiological signal fluctuation and increasing sensitivity for mapping.
This study compares rs-fMRI data preprocessing techniques in a patient with a brain tumor to investigate the effect of preprocessing on the data and different parameters, including mALFF, fALFF, ICA, and ReHo. The study also discusses the effect of preprocessing on the data and the four parameters means for the management and prognostication of brain tumor diagnosis and science.
Data and Methods
Data set information
96 subjects with primary brain tumors were selected to participate in the study. All participants provided informed written consent reviewed by their healthcare institutions. The description of the subject included the type of tumor, WHO grade, and scan duration using the rs-fMRI data preprocessing techniques. They include:
Group 1: glioblastoma, WHO Grade IV: 40 subjects with 28 male and 12 female of age range 23-77 diagnosed with new seizures and biopsy-proven left anterior. The scan duration was 10.04 minutes.
Group 2: anaplastic astrocytoma, WHO Grade III: 23 subjects, 15 male and 8 female of age range 24-71 with new-onset nocturnal seizure. The subjects were found to have a large tumor extending from the left frontal operculum to corona radiate, temporal operculum, and insula. The scan duration was 10.14 minutes.
Group 3: oligodendroglioma, WHO Grade II: 33 subjects, 18 male and 15 female of age range 19-73 experiencing first onset seizures. The subjects also experiencing speech component were established to have an extension of a non-contrast enhancing mass from the left frontal lobe to frontal operculum inferiorly and sensorimotor areas. The scan duration was 10.14 minutes.
Preprocessing
The preprocessing steps prior to subjecting raw fMRI data include:
Quality Assurance: The inspection of the source images was conducted together in montage mode to identify and exclude any form of individual slices that tend to appear too dark, too bright, or have artifacts.
Distortion Correction: Unwarping and field mapping were used to reduce any forms of distortions that may result from gradient echoes from the sr-fMRI sequences.
Slice Timing Correction: Data shifting strategy was utilized to reduce the slice timing differences. The strategy involves moving the recorded points to reflect their proper offset from the time of the stimulus.
Motion Correction: To avoid errors associated with head motion, the subjects underwent proper coaching and training prior to preprocessing. The head was also immobilized using padding and straps.
Temporal Preprocessing: To reduce the occurrence of fluctuations due to noise, a high-pass filtering process was conducted using time-domain averaging methods to eliminate detrending. For all subjects, probability maps for cerebrospinal fluid (CSF) and white matter (WM) were thresholded at p > 0.70 to create CSF and WM masks. Using these masks the BOLD time series was extracted from the resting state data set and the first five principal components were derived. We also included 24 motion parameters and motion scrubbing in to our temporal regression model. During temporal regression, we implemented a GLM (general linear model) based approaches where the BOLD fMRI signal from each voxel was treated as dependent variable and the nuisance covariates such as BOLD signal from CSF and WM motion were treated as independent variables. As the result of regression model, we derive the signal from the residual term, which implies part of BOLD fMRI signal that can not be explained by physiological noises represented by CSF and WM signal and motion related noises.

Spatial Smoothing: Reducing signal-to-noise ratio (SNR) resulting from spatial smoothing was conducted by multiplying the fMRI data with a 3D Gaussian filter (Vakamudi et al., 2020).
Table 1
Subject Grouping Based On Brain Tumor Progression
Grade Total No. of Subjects Male Female Age range
II 33 18 15 19 – 73
III 23 15 8 24 – 71
IV 40 28 12 23 – 77

Prior to statistical analysis, fMRI data undergoes a series of preprocessing steps in order to remove the artifacts and validate model assumptions. The preprocessing steps were the same for all subjects and we used SPM12 under MATLAB. We performed AC alignment to set the origin of all images to the Anterior Commissure location. All fMRI images undergo a motion correction to realign the time series of images to the mean of images for each subject. We derived 6 motion parameters that describe head motion in 6 directions XYZ (roll, pitch, yaw). Then we coregistered the fMRI images to the high resolution structural image. We performed segmentation on the structural images in order to extract gray matter, white matter, and CSF probability maps while deriving a deformation field. Finally, the fMRI data were transformed to MNI standard space using the deformation field derived from the segmentation step.

Figure 1. Preprocessing prior to statistical analysis
Table 2
Presurgical Resting‐State Functional Magnetic Resonance Imaging Scans
Subject Group Age/sex Tumor type rs-fMRI Scan duration (mm)
SG1

SG2

SG3

Age – 23-77
Female – 12
Male – 28

Age – 24-71
Female -8
Male – 15

Age – 19-73
Female -15
Male- 18

Glioblastoma WHO Grade IV

Anaplastic astrocytoma WHO Grade III

Oligodendroglioma WHO Grade II
10.04

10.14

10.14

Analysis
The rs-fMRI data preprocessing parameters that were utilized include mALFF, fALFF, ICA, and ReHo.
fALFF Analysis
fALFF analysis was conducted on voxel-wise amplitude of low-frequency fluctuation using AFNI’s 3dRSFC. To obtain the ALFF measures at each voxel, the calculation was conducted by first filtering the time series for each voxel to remove linear and quadratic trends through band-pass filtration (0.01-0.08 Hz). The filtered time series was then converted using a fast Fourier transform (FFT), allowing the power spectrum to be computed by squaring the amplitude at each frequency and obtaining the squire root of the power for each voxel in the range 0.01-0.1 HZ. The value was then divided by the global within-brain mean ALFF as a measure for every single voxel. The possibility of ALFF being influenced by physiological noise resulted in the examination of the fALFF to reduce the noise by calculating the fALFF in 3dRSFC as the ratio of the power in the low-frequency range relative to the whole frequency range. The fALFF at each voxel was then divided by whole-brain AFFF at each voxel to obtain the mAFLL. The extraction of the mAFLL and fALFF values from each hippocampal ROI was conducted, and Welch’s unpaired t-tests were used for analysis.
mALFF Analysis
Improving the original ALFF approach was conducted using a ratio of each frequency at a low-frequency range to that of the whole frequency range (0.01-0.08 HZ: 0.02 HZ). The sampling rate at each voxel was calculated using the AFNI’s 3dRSFCm, with the time series of each voxel converted by using a fast Fourier transform (FFT) and transformed to a frequency domain. The squire root at each frequency of the power spectrum was calculated for each voxel in the range of 0.01-0.08 HZ. To obtain the fALFF values at each voxel, the division of the sum of amplitude across 0.01-0.08 Hz by the amplitude across the entire frequency range (0.0.2 Hz) was calculated. The fALFF value allows the measurement of the relative contribution of specific LFO to the whole frequency range.
Independent Component Analysis (ICA)
The ICA was conducted at a group level and individual subject level. At a group level, the ICA was conducted using the Group ICA of fMRI Toolbox (McHugo et al., 2015). The calculation of the group-specific ICA was obtained by scanning time series to a mean of 100 on a voxel-wise basis. Complete spatiotemporal data set for each subject was then reduced to 35 principal components. The components were divided into three analyses of 20, 30, and 40 seconds, which were used to calculate the average hippocampal default mode network loadings and compare them with the independent components. At an individual subject’s level, the ICA analysis was conducted using FEAT and MELODIC tools. The preprocessing steps that were applied prior to the ICA analysis include normalization of the grand-mean intensity and variance, filtering high-pass temporal, and de-meaning voxel-wise. The performance of single-secession ICA was conducted multiple times on the fMRI time-series of each participant. The first single-secession ICA was carried out using the number of components that were estimated automatically by Bayesian model selection. The second single-secession ICA was carried out using the number of components specified. Each subject’s default mode network component was identified by selecting the unthresholded component map that had the highest correlation to the default mode network mask. The extraction of the mean independent component value from the hippocampal connectivity was conducted, and the value analyzed using the two-tailed, Welch’s unpaired t-tests.
Regional Homogeneity (ReHo) Analysis
ReHo was conducted to analyze the brain activity in the study subjects. ReHo is used in measuring of the spontaneous blood oxygenation level dependent (BOLD) fluctuations during rest, by targeting the local synchronization of voxels within a specific area of the brain. The analysis of the rs-fMRI data using ReHo involved measuring the indexes of locally synchronous brain activity by calculating the voxels fluctuations within specific voxel time-course. The calculation of ReHo was conducted based on the Kendall’s coefficient of concordance (KCC), whereby each voxel in the brain was calculated voxel-wise by apply a cluster size of 26 voxels basing on the formula below (Li et al., 2018).

Whereby, KCC of a specific voxels is W, ranging from 0-1, Ri was the rank sum of the ith time-point, K was the number of time-series within a measured cluster, and n was the number of ranks. The ReHo of each subject is obtained using the formula then the each subject ReHO map was divided by the global mean ReHo for standardization minimizing the influence of individual variations in the KCC value across the subjects.
Effect of Preprocessing On the Data and Different Parameters
The effect of preprocessing on the data and different parameters using the rs-fMRI parameters was observed in the study. In the fALFF analysis, the subjects were recorded to have increased amplitude of low frequency, which resulted in the increased connectivity between the subject’s BOLD signal fluctuations in a single voxel and all network regions. The increased amplitude of low frequency fastened the data acquisition process. The ReHo impact on data was observed through better measures that were obtained in the based connectivity analysis. The comparative analysis between different functional connectivity in the default mode network enhanced the elimination of errors in the data. The ReHo analysis had a significant impact on the rs-fMRI data, which was mainly associated with the brain connectivity and BOLD signal fluctuations. The ability of ReHo to reflect the synchronized activities in specific brain regions allowed efficient organization of data based on the information processing of the brain. The ICA analysis was significant in determining the impact of varying the number of independent components on the data. The connectivity at a trend level increased with the use of the ICA parameter, increasing the data visualization of the number of components greatly influenced by the variations of independent components. The sensitivity to changes in glutamatergic and GABAergic signals of the mALFF increased the data obtained due to the interictal epileptiform discharges.
Assignment help – Discussion about What This Means For Science
This study compared the Resting-state fMRI data preprocessing techniques in patients with brain tumors. The use of the rs-fMRI is considered to provide desirable scan time from the BOLD signals that can be used quickly, effectively, and credibly in detecting functional connectivity patterns. The management and prognostication of various types of brain tumors will improve using rs-fMRI data preprocessing techniques. The parameters used in rs-fMRI enable easy detection and recognition of various networks engaged during various cognitive tasks providing a boost to science. fALFF and mALFF ability to detect spontaneous activities across various states, such as when the subject is sleeping or under anesthesia, will significantly improve science in analyzing spontaneous fluctuations at rest related to a different memory, intrinsic processes, and thinking. The clinical study in various neurological and psychiatric disorders will be the most beneficiaries of the rs-fMRI data preprocessing techniques. The rs-fMRI has an improved data preprocessing ability compared to other methods and can operate without significant assist reducing the costs associated with its implementation. The science highly benefits from the rs-fMRI ability to obtain patient-specific diagnostic and prognostic information and predict neurological and psychiatric disorders.

References
Lee, M. H., Miller-Thomas, M. M., Benzinger, T. L., Marcus, D. S., Hacker, C. D., Leuthardt, E. C., & Shimony, J. S., 2016. Clinical Resting-state fMRI in the Preoperative Setting: Are We Ready for Prime Time?. Topics in magnetic resonance imaging : TMRI, 25(1), pp. 11–18. https://doi.org/10.1097/RMR.0000000000000075
Lin, J., Cui, X., Dai, X., & Mo, L., 2018. Regional Homogeneity Predicts Creative Insight: A Resting-State fMRI Study. Frontiers in human neuroscience, 12, 210. https://doi.org/10.3389/fnhum.2018.00210
Petrella, J., Shah, L., Harris, K., Friedman, A., George, T., Sampson, J., Pekala, J, & Voyvodic, J., 2006. Preoperative functional MR imaging localization of language and motor areas: effect on therapeutic decision making in patients with potentially resectable brain tumors. Radiology, 240(3), pp.793–802.
Sair, H. I., Yahyavi-Firouz-Abadi, N., Calhoun, V. D., Airan, R. D., Agarwal, S., Intrapiromkul, J., …Pillai, J. J., 2016. Presurgical brain mapping of the language network in patients with brain tumors using resting-state fMRI: Comparison with task fMRI. Human Brain Mapping, 37, 913–923.
Sharaev, M., Smirnov, A., Melnikova-Pitskhelauri, T., Orlov, V., Burnaev, E., Pronin, I., Pitskhelauri, D, & Bernstein, A., 2018. Functional Brain Areas Mapping in Patients with Glioma Based on Resting-State fMRI Data Decomposition. Conference: ICDM. doi:10.1109/ICDMW.2018.00049
Shimony J. S., Zhang, D., Johnston, J. M., Fox, M. D., Roy, A., & Leuthardt, E. C., 2009. Resting-state spontaneous fluctuations in brain activity: A new paradigm for presurgical planning using fMRI. Academic Radiology,16, 578–583.
Tie, Y., Rigolo, L., Norton, I. H., Huang, R. Y., Wu, W., Orringer, D., … Golby, A. J., 2013. Defining language networks from resting-state fMRI for surgical planning—A feasibility study. Human Brain Mapping, 35, 3683367.
Vakamudi, K., Posse, S., Jung, R., & Cushnyr, B., 2019. Real-time presurgical resting-state fMRI in patients with brain tumors: Quality control and comparison with task-fMRI and intraoperative mapping. Human Brain Mapping, 41(2), pp. 1-18.
Vakamudi, K., Posse, S., Jung, R., Cushnyr, B., & Chohan, M. O., 2020. Real-time presurgical resting-state fMRI in patients with brain tumors: Quality control and comparison with task-fMRI and intraoperative mapping. Human brain mapping, 41(3), pp. 797–814. https://doi.org/10.1002/hbm.24840
Zhang, D., Johnston, J. M., Fox, M. D., Leuthardt, E. C., Grubb, R. L., Chicoine, M. R., …Shimony, J. S., 2009. Preoperative sensorimotor mapping in brain tumor patients using spontaneous fluctuations in neuronal activity imaged with functional magnetic resonance imaging: Initial experience. Neurosurgery, 65, 226–236.
McHugo, M., Rogers, B. P., Talati, P., Woodward, N. D., & Heckers, S., 2015. Increased Amplitude of Low Frequency Fluctuations but Normal Hippocampal-Default Mode Network Connectivity in Schizophrenia. Frontiers in psychiatry, 6, 92. https://doi.org/10.3389/fpsyt.2015.00092

Order | Check Discount

Tags: Masters Essays, PSY Papers, PSYC, Psychology Assignment, Psychology Dissertations