Detection of physiological noise in resting state fMRI using machine learning

Reconstruction of respiratory variation signals from fMRI data

Functional MRI signals can be heavily influenced by systemic physiological processes in addition to local neural activity. For example, widespread hemodynamic fluctuations across the brain have been found to correlate with natural, low-frequency variations in the depth and rate of breathing over time. Acquiring peripheral measures of respiration during fMRI scanning not only allows for modeling such effects in fMRI analysis, but also provides valuable information for interrogating brain-body physiology. However, physiological recordings are frequently unavailable or have insufficient quality. Here, we propose a computational technique for reconstructing continuous low-frequency respiration volume (RV) fluctuations from fMRI data alone. We evaluate the performance of this approach across different fMRI preprocessing strategies. Further, we demonstrate that the predicted RV signals can account for similar patterns of temporal variation in resting-state fMRI data compared to measured RV fluctuations. These findings indicate that fluctuations in respiration volume can be extracted from fMRI alone, in the common scenario of missing or corrupted respiration recordings. The results have implications for enriching a large volume of existing fMRI datasets through retrospective addition of respiratory variations information.

Extraction of the cardiac waveform from simultaneous multislice fMRI data using slice sorted averaging and a deep learning reconstruction filter

Cardiac signal contamination has long confounded the analysis of blood-oxygenation-level-dependent (BOLD) functional magnetic resonance imaging (fMRI). Cardiac pulsation results in significant BOLD signal changes, especially in and around blood vessels. Until the advent of simultaneous multislice echo-planar imaging (EPI) acquisition, the time resolution of whole brain EPI was insufficient to avoid cardiac aliasing (and acquisitions with repetition times (TRs) under 400-500 ms are still uncommon). As a result, direct detection and removal of the cardiac signal with spectral filters is generally not possible. Modelling methods have been developed to mitigate cardiac contamination, and recently developed techniques permit the visualization of cardiac signal propagation through the brain in undersampled data (e.g., TRs > 1s), which is useful in its own right for finding blood vessels. However, both of these techniques require data from which to estimate cardiac phase, which is generally not available for the data in many large databases of existing imaging data, and even now is not routinely recorded in many fMRI experiments. Here we present a method to estimate the cardiac waveform directly from a multislice fMRI dataset, without additional physiological measurements, such as plethysmograms. The pervasive spatial extent and temporal structure of the cardiac contamination signal across the brain offers an opportunity to exploit the nature of multislice imaging to extract this signal from the fMRI data itself. While any particular slice is recorded at the TR of the imaging experiment, slices are recorded much more quickly - typically from 10 to 20 Hz - sufficiently fast to fully sample the cardiac signal. Using the fairly permissive assumptions that the cardiac signal is a) pseudoperiodic b) somewhat coherent within any given slice, and c) is similarly shaped throughout the brain, we can extract a good estimate of the cardiac phase as a function of time from fMRI data alone. If we make further assumptions about the shape and consistency of cardiac waveforms, we can develop a deep learning filter to greatly improve our estimate of the cardiac waveform.

Heart rate variability as a biomarker in health and affective disorders: A perspective on neuroimaging studies

The dynamic embodiment of psychological processes is evident in the association of health outcomes, behavioural traits and psychological functioning with Heart Rate Variability (HRV). The dominant high-frequency component of HRV is an index of the central neural control of heart rhythm, mediated via the parasympathetic vagus nerve. HRV provides a potential objective measure of action policies for the adaptive and predictive allostatic regulation of homeostasis within the cardiovascular system. In its support, a network of brain regions (referred to as the 'central autonomic network') maps internal state, and controls autonomic responses. This network includes regions of prefrontal cortex, anterior cingulate cortex, insula, amygdala, periaqueductal grey, pons and medulla. Human neuroimaging studies of neural activation and functional connectivity broadly endorse this architecture, and its link with cardiac regulation at rest and dysregulation in clinical states that include affective disorders. In this review, we appraise neuroimaging research and related evidence for HRV as an informative marker of autonomic integration with affect and cognition, taking a perspective on function and organisation. We consider evidence for the utility of HRV as a metric to inform targeted interventions to improve autonomic and affective dysregulation, and suggest research questions for further investigation.

Methods for cleaning the BOLD fMRI signal

Blood oxygen-level-dependent functional magnetic resonance imaging (BOLD fMRI) has rapidly become a popular technique for the investigation of brain function in healthy individuals, patients as well as in animal studies. However, the BOLD signal arises from a complex mixture of neuronal, metabolic and vascular processes, being therefore an indirect measure of neuronal activity, which is further severely corrupted by multiple non-neuronal fluctuations of instrumental, physiological or subject-specific origin. This review aims to provide a comprehensive summary of existing methods for cleaning the BOLD fMRI signal. The description is given from a methodological point of view, focusing on the operation of the different techniques in addition to pointing out the advantages and limitations in their application. Since motion-related and physiological noise fluctuations are two of the main noise components of the signal, techniques targeting their removal are primarily addressed, including both data-driven approaches and using external recordings. Data-driven approaches, which are less specific in the assumed model and can simultaneously reduce multiple noise fluctuations, are mainly based on data decomposition techniques such as principal and independent component analysis. Importantly, the usefulness of strategies that benefit from the information available in the phase component of the signal, or in multiple signal echoes is also highlighted. The use of global signal regression for denoising is also addressed. Finally, practical recommendations regarding the optimization of the preprocessing pipeline for the purpose of denoising and future venues of research are indicated. Through the review, we summarize the importance of signal denoising as an essential step in the analysis pipeline of task-based and resting state fMRI studies.

On the generalizability of resting-state fMRI machine learning classifiers

Machine learning classifiers have become increasingly popular tools to generate single-subject inferences from fMRI data. With this transition from the traditional group level difference investigations to single-subject inference, the application of machine learning methods can be seen as a considerable step forward. Existing studies, however, have given scarce or no information on the generalizability to other subject samples, limiting the use of such published classifiers in other research projects. We conducted a simulation study using publicly available resting-state fMRI data from the 1000 Functional Connectomes and COBRE projects to examine the generalizability of classifiers based on regional homogeneity of resting-state time series. While classification accuracies of up to 0.8 (using sex as the target variable) could be achieved on test datasets drawn from the same study as the training dataset, the generalizability of classifiers to different study samples proved to be limited albeit above chance. This shows that on the one hand a certain amount of generalizability can robustly be expected, but on the other hand this generalizability should not be overestimated. Indeed, this study substantiates the need to include data from several sites in a study investigating machine learning classifiers with the aim of generalizability.

Highcor: a novel data-driven regressor identification method for BOLD fMRI

A data-driven regressor selection technique for reducing physiological noise in BOLD fMRI is presented that capitalizes on additional information contained in the phase of the signal time-course. This method, termed highcor, identifies a set of suspect voxels by selecting based on high temporal correlation between the magnitude and phase components of the time-course. Temporal regressors are generated from principal component analysis of this voxel set. Regressor spectral content is investigated with high temporal resolution datasets, and filtering performance is demonstrated. The technique is benchmarked against compcor, an increasingly popular data driven technique. Highcor was found to select a unique set of physiological noise source voxels, and identification of confound and physiologically related signals was robust even at slow temporal sampling rates. Filtering using regressors derived from compcor and highcor voxels resulted in reductions in overall temporal standard deviation in cortical areas of 16.1%±3.1%, and 18.1%±3.8%, (mean±sd.) as measured over 36 BOLD fMRI datasets that featured an anti-saccade task. An approach combining both methods resulted in further reductions of temporal standard deviations by 31.4%±3.8%. In these regions, mean temporal SNR values were improved from 38.7±3.4 to 47.7±3.7 (cc), 49.2±2.2 (hc), and 57.8±2.3 (hc+cc). tSNR increases from data-driven filtering translated into some associated improvements in overall detection of task in the sample datasets.