Adaptive filtering to reduce global interference in non-invasive NIRS measures of brain activation: how well and when does it work?

Department of Psychiatry, Massachusetts General Hospital, Harvard Medical School, Charlestown, MA 02129, USA.
NeuroImage (Impact Factor: 6.13). 02/2009; 45(3):788-94. DOI: 10.1016/j.neuroimage.2008.12.048
Source: PubMed

ABSTRACT In previous work we introduced a novel method for reducing global interference, based on adaptive filtering, to improve the contrast to noise ratio (CNR) of evoked hemodynamic responses measured non-invasively with near infrared spectroscopy (NIRS). Here, we address the issue of how to generally apply the proposed adaptive filtering method. A total of 156 evoked visual response measurements, collected from 15 individuals, were analyzed. The similarity (correlation) between measurements with far and near source-detector separations collected during the rest period before visual stimulation was used as indicator of global interference dominance. A detailed analysis of CNR improvement in oxy-hemoglobin (O(2)Hb) and deoxy-hemoglobin (HHb), as a function of the rest period correlation coefficient, is presented. Results show that for O(2)Hb measurements, 66% exhibited substantial global interference. For this dataset, dominated by global interference, 71% of the measurements revealed CNR improvements after adaptive filtering, with a mean CNR improvement of 60%. No CNR improvement was observed for HHb. This study corroborates our previous finding that adaptive filtering provides an effective method to increase CNR when there is strong global interference, and also provides a practical way for determining when and where to apply this technique.

  • [Show abstract] [Hide abstract]
    ABSTRACT: Functional near-infrared spectroscopy (fNIRS) is an increasingly popular technology for studying brain function because it is non-invasive, non-irradiating and relatively inexpensive. Further, fNIRS potentially allows measurement of hemodynamic activity with high temporal resolution (milliseconds) and in naturalistic settings. However, in comparison with other imaging modalities, namely fMRI, fNIRS has a significant drawback: limited sensitivity to hemodynamic changes in deep-brain regions. To overcome this limitation, we developed a computational method to infer deep-brain activity using fNIRS measurements of cortical activity. Using simultaneous fNIRS and fMRI, we measured brain activity in 17 participants as they completed three cognitive tasks. A support vector regression (SVR) learning algorithm was used to predict activity in twelve deep-brain regions using information from surface fNIRS measurements. We compared these predictions against actual fMRI-measured activity using Pearson’s correlation to quantify prediction performance. To provide a benchmark for comparison, we also used fMRI measurements of cortical activity to infer deep-brain activity. When using fMRI-measured activity from the entire cortex, we were able to predict deep-brain activity in the fusiform cortex with an average correlation coefficient of 0.80 and in all deep-brain regions with an average correlation coefficient of 0.67. The top 15% of predictions using fNIRS signal achieved an accuracy of 0.7. To our knowledge, this study is the first to investigate the feasibility of using cortical activity to infer deep-brain activity. This new method has the potential to extend fNIRS applications in cognitive and clinical neuroscience research.
    Biomedical Optics Express 03/2015; 6(3). DOI:10.1364/BOE.6.001074 · 3.50 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: brain-computer interface (BCI) is a communication system that allows the use of brain activity to control computers or other external devices. It can, by bypassing the peripheral nervous system, provide a means of communication for people suffering from severe motor disabilities or in a persistent vegetative state. In this paper, brain-signal generation tasks, noise removal methods, feature extraction/selection schemes, and classification techniques for fNIRS-based BCI are reviewed. The most common brain areas for fNIRS BCI are the primary motor cortex and the prefrontal cortex. In relation to the motor cortex, motor imagery tasks were preferred to motor execution tasks since possible proprioceptive feedback could be avoided. In relation to the prefrontal cortex, fNIRS showed a significant advantage due to no hair in detecting the cognitive tasks like mental arithmetic, music imagery, emotion induction, etc. In removing physiological noise in fNIRS data, band-pass filtering was mostly used. However, more advanced techniques like adaptive filtering, independent component analysis, multi optodes arrangement, etc. are being pursued to overcome the problem that a band-pass filter cannot be used when both brain and physiological signals occur within a close band. In extracting features related to the desired brain signal, the mean, variance, peak value, slope, skewness, and kurtosis of the noised-removed hemodynamic response were used. For classification, the linear discriminant analysis method provided simple but good performance among others: support vector machine, hidden Markov model, artificial neural network, etc. fNIRS will be more widely used to monitor the occurrence of neuro-plasticity after neuro-rehabilitation and neuro-stimulation. Technical breakthroughs in the future are expected via bundled-type probes, hybrid EEG-fNIRS BCI, and through the detection of initial dips.
    Frontiers in Human Neuroscience 01/2015; DOI:10.3389/fnhum.2015.00003 · 2.90 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: Objective. Functional near-infrared spectroscopy (fNIRS) is an emerging technique for the in vivo assessment of functional activity of the cerebral cortex as well as in the field of brain–computer interface (BCI) research. A common challenge for the utilization of fNIRS in these areas is a stable and reliable investigation of the spatio-temporal hemodynamic patterns. However, the recorded patterns may be influenced and superimposed by signals generated from physiological processes, resulting in an inaccurate estimation of the cortical activity. Up to now only a few studies have investigated these influences, and still less has been attempted to remove/reduce these influences. The present study aims to gain insights into the reduction of physiological rhythms in hemodynamic signals (oxygenated hemoglobin (oxy-Hb), deoxygenated hemoglobin (deoxy-Hb)). Approach. We introduce the use of three different signal processing approaches (spatial filtering, a common average reference (CAR) method; independent component analysis (ICA); and transfer function (TF) models) to reduce the influence of respiratory and blood pressure (BP) rhythms on the hemodynamic responses. Main results. All approaches produce large reductions in BP and respiration influences on the oxy-Hb signals and, therefore, improve the contrast-to-noise ratio (CNR). In contrast, for deoxy-Hb signals CAR and ICA did not improve the CNR. However, for the TF approach, a CNR-improvement in deoxy-Hb can also be found. Significance. The present study investigates the application of different signal processing approaches to reduce the influences of physiological rhythms on the hemodynamic responses. In addition to the identification of the best signal processing method, we also show the importance of noise reduction in fNIRS data.
    Journal of Neural Engineering 08/2014; 11(5):056010. DOI:10.1088/1741-2560/11/5/056010 · 3.42 Impact Factor

Full-text (2 Sources)

Available from
May 30, 2014