Automatic denoising of functional MRI data: combining independent component analysis and hierarchical fusion of classifiers

Abstract

Many sources of fluctuation contribute to the fMRI signal, and this makes identifying the effects that are truly related to the underlying neuronal activity difficult. Independent component analysis (ICA) - one of the most widely used techniques for the exploratory analysis of fMRI data - has shown to be a powerful technique in identifying various sources of neuronally-related and artefactual fluctuation in fMRI data (both with the application of external stimuli and with the subject "at rest"). ICA decomposes fMRI data into patterns of activity (a set of spatial maps and their corresponding time series) that are statistically independent and add linearly to explain voxel-wise time series. Given the set of ICA components, if the components representing "signal" (brain activity) can be distinguished form the "noise" components (effects of motion, non-neuronal physiology, scanner artefacts and other nuisance sources), the latter can then be removed from the data, providing an effective cleanup of structured noise. Manual classification of components is labour intensive and requires expertise; hence, a fully automatic noise detection algorithm that can reliably detect various types of noise sources (in both task and resting fMRI) is desirable. In this paper, we introduce FIX ("FMRIB's ICA-based X-noiseifier"), which provides an automatic solution for denoising fMRI data via accurate classification of ICA components. For each ICA component FIX generates a large number of distinct spatial and temporal features, each describing a different aspect of the data (e.g., what proportion of temporal fluctuations are at high frequencies). The set of features is then fed into a multi-level classifier (built around several different classifiers). Once trained through the hand-classification of a sufficient number of training datasets, the classifier can then automatically classify new datasets. The noise components can then be subtracted from (or regressed out of) the original data, to provide automated cleanup. On conventional resting-state fMRI (rfMRI) single-run datasets, FIX achieved about 95% overall accuracy. On high-quality rfMRI data from the Human Connectome Project, FIX achieves over 99% classification accuracy, and as a result is being used in the default rfMRI processing pipeline for generating HCP connectomes. FIX is publicly available as a plugin for FSL.

Figure 1

Examples of “good” components from three different acquisitions. The spatial map for the high resolution, short TR acquisition (bottom; acquisition C, see text for more details) is visually strikingly different from a more standard acquisition (top and middle, acquisitions A and B, see text), with the signal above threshold following very closely the cortical gyrification. The spectral power lies primarily between 0 and 0.05 Hz for each component.

Figure 2

Example movement-related artefacts. The signal above threshold in the spatial maps is essentially at the edges of the brain. The frequencies of the power spectra are disparately distributed and the time courses visually dissimilar.

Figure 3

Two further noise components: “white matter” and “susceptibility-motion”.

Figure 4

Examples of cardiac-related components. This includes components due to cardiac pulsation and arterial contribution. The signal above threshold in the spatial maps is essentially located in the ventricles, or following the main arteries (posterior cerebral artery, middle cerebral branches).

Figure 5

Example components relating to large veins. The signal above threshold in the spatial maps is essentially following the sagittal sinus.

Figure 6

Two MRI acquisition/reconstruction related artefact components.

Figure 7

Two examples of “unknown” components.

Figure 8

FIX’s hierarchical classifier. In the data layer, full, feature-selected, temporal, spatial, temporal-feature-selected and spatial-feature-selected datasets ( , , , , and , respectively), are each classified by 5 Classifiers. These Classifiers consist of k-NN, SVM_r (SVM with RBF kernel), SVM_p (SVM with polynomial kernel), SVM_l (linear SVM) and decision tree (simply called tree here). The result is a vector of 30 (5 × 6) probabilities (0 and 1 denoting perfect noise and perfect signal, respectively), which is the input to a fusion-layer classifier, whose output is the probability of IC being signal/noise.

Figure 9

FIX-RF and FIX-SVM-RBF outperform the commonly-used classifiers on a broad set of rFMRI datasets that cover a board range of data-acquisition/-quality scenarios, common in rFMRI. In the figure, classifiers are shown on the x-axis, and the y-axis shows the average accuracy across all datasets. For each dataset, accuracy is defined as the average of subject-wise (TPR+TNR)/2 (see Section 2.5), where TPR and TNR denote the true positive and true negative rates, respectively. The thick blue and red lines show the mean and median of accuracy across datasets, respectively, and dashed blue and red lines shows the best classifier’s (i.e., FIX-RF) performance in terms of its mean and median, respectively. Thus, on average, FIX is expected to outperform other classifiers, and the best simple classifier next to FIX is SVM-RBF.