Model Independent fMRI Data Analyses fMRI Methods Journal Club Andy James Tuesday, March 8, 2005
Independent Analyses Advantages • Require no a priori hypotheses • Can capture influences not relating to model Disadvantages • Computationally/statistically complex • Less intuitive than model-dependent methods • Less specificity for cluster function
Methods to Discuss • Principle Component Analysis (PCA) • Independent Component Analysis (ICA) • Temporal Clustering Analysis (TCA) • Connectivity / Correlational Analyses [specifically Within-condition Interregional Covariate Analysis (WICA)]
PCAICATCAWICA Conceptualizing PCA
PCAICATCAWICA Conceptualizing PCA brain’s total spatial and temporal variance variance of voxels’ individual timecourses =
PCAICATCAWICA Conceptualizing PCA brain’s total spatial and temporal variance variance of voxels’ individual timecourses = BUT, some voxels better explain overall variance than others! PCA asks: How can we cluster voxels into components to best explain the brain’s variance?
PCAICATCAWICA PCA Visuomotor example Visual Stimuli Subject Response time Some brain regions (V1, M1, cerebellum, thalamus, SMA) should have greater temporal variability (more variance) than others (Broca’s area, sylvian fissure, amygdala, etc.)
PCAICATCAWICA Conceptualizing PCA Component 1: Visual cortex Comp 1 Component 2: Motor system Comp 2 Comp 1 Comp 3 Component 3: Frontal-visual Comp 2 Comp 1
PCAICATCAWICA PCA limitations • “If task-related fMRI changes are only a small part of total signal variance… capturing the greatest variance in the data may reveal little information about task-related activations.” (McKeown, 1998) • PCA also necessitates orthogonal components– i.e. no interactions across defined systems • Ex: view pictures of horror, disgust, sex & bunnies • PCA cannot account for differential task-induced activity of amygdala, V1, insula, and prefrontal cortex
PCAICATCAWICA Independent Component Analysis • Related to PCA, ICA deconvolves a mixture of signals into sources. • Widely accepted as more powerful and sensitive than PCA. • Matlab: FastICA (McKeown, 1998)
PCAICATCAWICA Another ICA illustration (McKeown, 1998)
PCAICATCAWICA Conceptualizing ICA (McKeown, 1998)
PCAICATCAWICA ICA Comparisons 3 participants performed the Stroop test. ICA yielded multiple components; including one whose timecourse closely matched the paradigm (shown right) (McKeown, 1998)
PCAICATCAWICA (McKeown, 1998)
PCAICATCAWICA ICA Comparisons (McKeown, 1998)
PCAICATCAWICA ICA Comparisons An additional participant performed a word/number task: Participant viewed a word (2s) followed by mental arithmetic (6s). (still McKeown, 1998)
PCAICATCAWICA Additional comments A voxel can contribute to multiple components. ICA reveals non-task specific components. ICA could be valuable for masking unwanted voxels (i.e. slowly-varying) (Guess)
PCAICATCAWICA Temporal Clustering Analysis • Useful if expecting brain response to stimulus but do not know when it will occur • Plot the maximum voxel intensity for each timepoint Adapted from Liu et al., 2000.
PCAICATCAWICA Temporal Clustering Analysis • Best used for detecting robust response to a single stimulus (ex: glucose consumption) • Often a precursor to model-dependent analyses • Limited utility when dealing with multiple or concurrent processes, weak task-related neural responses, or noise from a continuous process (i.e. motor response)
PCAICATCAWICA Connectivity Analyses TO BE CONTINUED…
References • McKeown MJ, Makeig S, Brown GG, Jung T-P, Kindermann SS, Bell AJ & Sejnowski TJ. (1998). Analysis of fMRI data by blind separation into independent spatial components. Human Brain Mapping, 6, 160-188. • Liu Y, Gao J-H, Liu H-L & Fox PT. (2000). The temporal response of the brain after eating revealed by functional MRI. Nature, 405, 1058-1062.