Task-related component analysis for functional neuroimaging and application to near-infrared spectroscopy data

Reproducibility of experimental results lies at the heart of scientific disciplines. Here we propose a signal processing method that extracts task-related components by maximizing the reproducibility during task periods from neuroimaging data. Unlike hypothesis-driven methods such as general linear models, no specific time courses are presumed, and unlike data-driven approaches such as independent component analysis, no arbitrary interpretation of components is needed. Task-related components are constructed by a linear, weighted sum of multiple time courses, and its weights are optimized so as to maximize inter-block correlations (CorrMax) or covariances (CovMax). Our analysis method is referred to as task-related component analysis (TRCA). The covariance maximization is formulated as a Rayleigh-Ritz eigenvalue problem, and corresponding eigenvectors give candidates of task-related components. In addition, a systematic statistical test based on eigenvalues is proposed, so task-related and -unrelated components are classified objectively and automatically. The proposed test of statistical significance is found to be independent of the degree of autocorrelation in data if the task duration is sufficiently longer than the temporal scale of autocorrelation, so TRCA can be applied to data with autocorrelation without any modification. We demonstrate that simple extensions of TRCA can provide most distinctive signals for two tasks and can integrate multiple modalities of information to remove task-unrelated artifacts. TRCA was successfully applied to synthetic data as well as near-infrared spectroscopy (NIRS) data of finger tapping. There were two statistically significant task-related components; one was a hemodynamic response, and another was a piece-wise linear time course. In summary, we conclude that TRCA has a wide range of applications in multi-channel biophysical and behavioral measurements.