Functional magnetic resonance imaging (fMRI) has become a predominant modality to investigate neuronal activation patterns in human brain due to its relatively low invasiveness. There are various analytical options to estimate neuronal activation patterns from fMRI data. In this thesis, there are su...
Functional magnetic resonance imaging (fMRI) has become a predominant modality to investigate neuronal activation patterns in human brain due to its relatively low invasiveness. There are various analytical options to estimate neuronal activation patterns from fMRI data. In this thesis, there are suggestions of two analytical options to better estimate individual-level neuronal activation patterns via a novel iterative alternating least-squares based approach and to better estimate reliable group-level neuronal activation patterns of default-mode networks (DMNs) via group-statistics.
The proposed analytical option for individual-level is iterative dual-regression with a sparse prior approach. Among available analytical options to estimate individual-level neuronal activation patterns, dual-regression (DR) can be utilized based on alternating least-squares (ALS) algorithm using group-level neuronal activation patterns from group-level analysis. However, original DR performs the ALS at once to estimate individual-level patterns, in this respect, the estimated individual-level activation patterns may not enough to explain the individual’s characteristics. Also, the estimated individual-level activation patterns from original DR have been claimed that can potentially be sensitive to individual noises. To resolve this problem, iterative gradient updating and increasing sparseness of spatial patterns were additionally considered in the proposed iterative DR approach. To quantitatively evaluate the performance of proposed approach, spatially- and temporally- varied semi-artificial fMRI data were used. Also, to show the potential ability of practical utilization of proposed approach, simple motor task based real fMRI data were used. To show the improved performances of proposed iterative DR approach, general linear model (GLM) and original DR approach were used among available analytical options to estimate individual-level neuronal activation patterns. From the measured performances via receiver operating characteristic power and Pearson’s correlation coefficients, proposed DR showed better estimation performances than original DR, as expected. Based on evidences of greater performances, proposed iterative DR with a sparse prior can strongly be suggested to investigate individual-level neuronal activation patterns.
The suggested analytical option for group-level is a mixed-effect analysis (MFX) than mostly utilized random-effect analysis (RFX) for the DMN study. The RFX statistic estimates the group-level neuronal activation patterns with only considering inter-subject variability based on Student’s t-test statistic. In this respect, RFX can potentially be vulnerable to a tested group if there are substantial levels of variability of activation patterns for each subject. To prevent a potentially significant level of variability in group-level spatial patterns, MFX can be better optimized analytical option to estimate group-level spatial patterns with considering intra-subject variability as well as inter-subject variability. Through the previous studies showed the greater level of reproducibility in MFX statistic than RFX statistic using task-based fMRI data, however, quantitative evaluation of reproducibility between group statistics was not performed on resting-state fMRI data. In this thesis, the analytical option of MFX statistic was performed to resting-state fMRI data with estimation of intra-subject variability via the dual-regression. Group-level default-mode networks (DMNs), the part of resting-state networks (RSNs), were estimated via two options of group statistic, then, group statistics were assessed in terms of reproducibility and variability based on bootstrapping evaluation. The adopted performance measures to quantitatively compare the reproducibility of group statistic were (1) neuronal activation level in the DMNs of group spatial patterns, (2) Pearson’s spatial correlation coefficients between two subgroups from a group statistic, (3) percentage of overlapped active clusters between two subgroups, and (4) distances of center-of-clusters between subgroups. As expected, MFX showed greater level of reproducibility than RFX on DMNs. Additionally, regional characteristics of DMNs were observed greater level of neuronal activations in posterior part than anterior part of DMNs and greater degrees of reproducibility in medial parts than lateral part of DMNs. These findings may provide a valuable option to investigate neuropsychiatric characteristic traits using DMNs study.
Functional magnetic resonance imaging (fMRI) has become a predominant modality to investigate neuronal activation patterns in human brain due to its relatively low invasiveness. There are various analytical options to estimate neuronal activation patterns from fMRI data. In this thesis, there are suggestions of two analytical options to better estimate individual-level neuronal activation patterns via a novel iterative alternating least-squares based approach and to better estimate reliable group-level neuronal activation patterns of default-mode networks (DMNs) via group-statistics.
The proposed analytical option for individual-level is iterative dual-regression with a sparse prior approach. Among available analytical options to estimate individual-level neuronal activation patterns, dual-regression (DR) can be utilized based on alternating least-squares (ALS) algorithm using group-level neuronal activation patterns from group-level analysis. However, original DR performs the ALS at once to estimate individual-level patterns, in this respect, the estimated individual-level activation patterns may not enough to explain the individual’s characteristics. Also, the estimated individual-level activation patterns from original DR have been claimed that can potentially be sensitive to individual noises. To resolve this problem, iterative gradient updating and increasing sparseness of spatial patterns were additionally considered in the proposed iterative DR approach. To quantitatively evaluate the performance of proposed approach, spatially- and temporally- varied semi-artificial fMRI data were used. Also, to show the potential ability of practical utilization of proposed approach, simple motor task based real fMRI data were used. To show the improved performances of proposed iterative DR approach, general linear model (GLM) and original DR approach were used among available analytical options to estimate individual-level neuronal activation patterns. From the measured performances via receiver operating characteristic power and Pearson’s correlation coefficients, proposed DR showed better estimation performances than original DR, as expected. Based on evidences of greater performances, proposed iterative DR with a sparse prior can strongly be suggested to investigate individual-level neuronal activation patterns.
The suggested analytical option for group-level is a mixed-effect analysis (MFX) than mostly utilized random-effect analysis (RFX) for the DMN study. The RFX statistic estimates the group-level neuronal activation patterns with only considering inter-subject variability based on Student’s t-test statistic. In this respect, RFX can potentially be vulnerable to a tested group if there are substantial levels of variability of activation patterns for each subject. To prevent a potentially significant level of variability in group-level spatial patterns, MFX can be better optimized analytical option to estimate group-level spatial patterns with considering intra-subject variability as well as inter-subject variability. Through the previous studies showed the greater level of reproducibility in MFX statistic than RFX statistic using task-based fMRI data, however, quantitative evaluation of reproducibility between group statistics was not performed on resting-state fMRI data. In this thesis, the analytical option of MFX statistic was performed to resting-state fMRI data with estimation of intra-subject variability via the dual-regression. Group-level default-mode networks (DMNs), the part of resting-state networks (RSNs), were estimated via two options of group statistic, then, group statistics were assessed in terms of reproducibility and variability based on bootstrapping evaluation. The adopted performance measures to quantitatively compare the reproducibility of group statistic were (1) neuronal activation level in the DMNs of group spatial patterns, (2) Pearson’s spatial correlation coefficients between two subgroups from a group statistic, (3) percentage of overlapped active clusters between two subgroups, and (4) distances of center-of-clusters between subgroups. As expected, MFX showed greater level of reproducibility than RFX on DMNs. Additionally, regional characteristics of DMNs were observed greater level of neuronal activations in posterior part than anterior part of DMNs and greater degrees of reproducibility in medial parts than lateral part of DMNs. These findings may provide a valuable option to investigate neuropsychiatric characteristic traits using DMNs study.
주제어
#resting-state
#fMRI
※ AI-Helper는 부적절한 답변을 할 수 있습니다.