Independent component analysis (ICA) is the most commonly used computational tool for identifying and characterizing underlying brain functional networks. We have done pioneering work in developing formal statistical modeling, estimation and inference framework for ICA in neuroimaging studies, particularly in the area of investigating between-subject heterogeneity in ICA-derived brain networks and modeling longitudinal changes in brain functional networks.
General temporal-concatenation group ICA (TC-GICA) models that can accommodate different types of between-subject variability in temporal responses and model wide variety of neural signals under different experimental tasks (Guo and Pagnoni, 2008; Guo, Biometrics, 2011).
A hierarchical ICA framework. we have developed a novel hierarchical ICA framework to formally model subject-specific effects in both temporal and spatial domains in fMRI data (Guo and Tang, 2013; Shi and Guo, 2016). In particular, the hierarchical covariate-adjusted ICA (hc-ICA) model (Shi and Guo, 2016) provides the first group ICA method that allows formally modeling and testing of covariate effects in ICA decomposition. Our method provides a more reliable and powerful statistical tool for evaluating group differences in brain functional networks while appropriately controlling for potential confounding factors. We have released a Matlab GUI-based toolbox HINT for implementing the hc-ICA.
Longitudinal ICA. We have developed a longitudinal ICA (L-ICA) model (Wang and Guo, 2019) to investigate neurodevelopment-, disease- or treatment-related changes in brain networks. With existing ICA methods only applicable to cross-sectional imaging studies, our proposed L-ICA helps fill a major gap in ICA methodology.
SparseBayes ICA. We have developed Sparse Bayesian Independent Component Analysis (SparseBayes ICA) for reliable estimation of individual differences in brain networks. SparseBayes ICA presents a powerful approach for modeling brain network source signals using a flexible Dirichlet process mixture, coupled with sparse estimation of covariate effects via a horseshoe prior. This advanced methodology significantly enhances the accuracy of uncovering latent brain networks and also improves both the sensitivity and specificity in detecting individual differences on brain networks.
References
†: denote students/postdocs advised/co-advised
Lukemire, J.†, Pagnoni, G. and Guo, Y. (2023). Sparse Bayesian Modeling of Hierarchical ICA: Reliable Estimation of Individual Differences in Brain Networks. Biometrics. In Press
Wang, Y. † and Guo, Y (2019). A hierarchical independent component analysis model for longitudinal neuroimaging studies. NeuroImage, 189(1), 380-400.
Shi, R.† and Guo, Y (2016). Modeling Covariate Effects in Group Independent Component Analysis with Application to Functional Magnetic Resonance Imaging. Annals of Applied Statistics. (Accepted). An earlier version of the paper was selected for Best Student Paper Award, American Statistical Association (ASA) Statistics in Imaging Section.
Guo, Y and Tang, Li. (2013). A hierarchical probabilistic model for group independent component analysis in fMRI studies, Biometrics, doi: 10.1111/biom.12068.
Guo Y (2011). A general probabilistic model for group independent component analysis and its estimation methods. Biometrics. 67(4): 1532-1542.
Guo Y and Pagnoni G (2008). A unified framework for group independent component analysis for multi-subject fMRI data. NeuroImage 42: 1078-1093. Listed in ScienceDirect’s Top 25 hottest articles of NeuroImage between July-Sep. 2008.
Guo Y (2008). Group Independent Component Analysis of Multi-subject fMRI data: Connections and Distinctions between Two Methods. IEEE Proceedings of the 2008 International Conference on BioMedical Engineering and Informatics v2: 748-752.
We have developed innovative blind source separation approaches for decomposing brain connectivity data to reveal latent connectivity source signals and characterize individuals' loadings on these sources.
References
We have developed a variety of network analysis tools to generate more reliable and reproducible results for investigating brain network and connectivity. These tools derive from a variety of advanced statistical methodologies including sparse inverse convariance estimation and graphical models.
1. Direct brain connectivity via partial correlations. We have developed an efficient and reliable statistical method for estimating direct functional connectivity in large-scale brain networks via partial correlations (Wang et al., 2016). Based on this work, we have released an R package DensParcorr.
2. New Graphical Model Approaches. We have developed novel graphical model approaches of brain network research including: (a) dynamic brain connectivity (Kundu et al., 2018, NeuroImage), (b) integrative analysis of brain networks by fusing functional MRI and diffusion MRI (Higgins et al., 2018, NeuroImage) and (c) joint analysis of multiple brain networks using both task-related and resting state fMRI data (Lukemire et al 2019+, JASA).
References
Lukemire, J†, Kundu, S, Pagnoni, G and Guo, Y * (2019+). An integrative Bayesian approach for joint modeling of multiple brain networks, JASA, Accepted with minor revisions.
Higgins, IA†, Kundu, S. and Guo, Y., (2018). Integrative Bayesian Analysis of Brain Functional Networks Incorporating Anatomical Knowledge. NeuroImage, 181: 263-278.
Kundu, S., Ming, J., Pierce, J. McDowell, J. and Guo, Y *, (2018), Estimating Dynamic Brain Functional Networks Using Multi-subject fMRI Data, NeuroImage, 183: 635-649.
Wang, Y†, Kang J, Kemmer, PB and Guo, Y*. (2016). An efficient and reliable statistical method for estimating functional connectivity in large-scale brain networks using partial correlation. Frontiers in Neuroscience. 10:123. doi: 10.3389/fnins.2016.00123.
We have developed statistical models to predict subjects’ clinical outcomes and cognitive states based on brain images. The proposed methods contribute to enhancing the utility of neuroimaging techniques in clinical practice, particularly in developing personalized interventions for more effective treatment of mental disorders.
1. A predictive model for forecasting future neural activation. This predictive model has potentials in helping select optimal treatment plans for individual patients in clinical practice. For example, it has been successfully applied in a neuroimaging study of schizophrenia to predict patients’ post-treatment neural activity after various treatments (Guo et al., Human Brain Mapping, 2008).
2. A weighted cluster kernel PCA model for predicting subjects’ cognitive states using brain images (Guo, 2010). The kernel PCA model improves prediction accuracy by capturing nonlinearity patterns in images and is robust to outlying subjects.
3. A Bayesian hierarchical model to provide predictive information on disease progression- or treatment-related changes in brain connectivity. The proposed method improves the accuracy of individualized prediction of connectivity by combining information from both group-level connectivity patterns that are common across subjects as well as unique individual-level connectivity features. Our predictive model has been applied to Alzheimer's Disease Neuroimaging Initiative (ADNI) data to predict longitudinal changes in neural circuits in Alzheimer patients (Dai and Guo, NeuroImage, 2017).
References
Dai, T† and Guo, Y * (2017). Predicting Individual Brain Functional Connectivity Using a Bayesian Hierarchical Model. NeuroImage.147(15): 772–787.
Guo Y (2010). A weighted cluster kernel PCA prediction model for multi-subject brain imaging data. Statistics And Its Interface. 3:103-111.
Guo Y, Bowman FD, Kilts C (2008). Predicting the brain response to treatment using a Bayesian Hierarchical model . Human Brain Mapping, 29(9): 1092-1109.
1. Agreement Methods for Time-to-Event Data Agreement studies have wide and important applications in biomedical research and clinical practices since multiple-raters/methods measurements commonly occur in such settings. One challenging topic in agreement studies is how to account for censored or truncated observations such as those observed in survival studies. Our work in this area includes: 1) development of one of the first nonparametric estimation methods for agreement measures in the presence of censoring (Guo and Manatunga, 2007 and 2009), 2) development of agreement methods to accommodate different types of survival outcomes including both discrete (Guo and Manatunga, 2005 and 2009) and continuous data (Guo and Manatunga, 2007 and 2010); 3) development of modeling tools in agreement studies that allow for modeling subjects’ covariate effects on the strength of agreement (Guo and Manatunga, 2005). Recently, we have developed a novel framework for assessing agreement based on survival processes (Guo et al., 2013). This new framework circumvents the need of estimating moment functions of survival times in the previous agreement methods and can be widely applied to survival studies with various study lengths.
2. Agreement Methods for Multi-scale and High Dimensional Data We have developed new statistical methods that extend the existing agreement paradigm to handle multiple scale (continuous/ordinal) scenarios. Additionally, we have worked on development of new agreement methodology to investigate the alignment between traditional behavior/clinical outcomes and neuroimaging biomarkers and to assess agreement between images acquired from multi-center neuroimaging studies.
References
Guo Y, Li R, Peng L and Manatunga AK. (2013). A new agreement measures based on survival processes, Biometrics, doi: 10.1111/biom.12063.
Peng L, Li R, Guo Y, and Manatunga AK.(2011). Assessing Broad Sense Agreement between Ordinal and Continuous Measurements. Journal of American Statistician Association. 106: 1592-1601. Selected as JASA Featured Article.
Guo Y and Manatunga AK (2010). A note on assessing agreement for frailty models. Statistics and Probability Letters. 80: 527-533.
Guo Y and Manatunga AK (2009). Measuring agreement of multivariate discrete survival times using a modified weighted kappa coefficient. Biometrics, 65(1):125-34.
Guo Y and Manatunga AK (2007). Nonparametric estimation of the concordance correlation coefficient under univariate censoring. Biometrics, 63(1): 164-172.
Guo Y and Manatunga AK (2005). Modeling the agreement of discrete bivariate survival times using kappa coefficient. Lifetime Data Analysis, 11(3): 309-332