A new approach to estimating the evoked hemodynamic response applied to dual channel functional near infrared spectroscopy

Highlights

• We present a novel algorithm to extract the brain hemodynamic response from the fNIRS signals.

• fNIRS signals with non-stationary property of physiological components were simulated.

• The proposed method does not require a prior assumption on the amplitude, shape and duration of the hemodynamic responses.

• The proposed method has significant advantages that make it appropriate for real-time applications.

Abstract

Background

Brain activity can be measured non-invasively by means of functional near-infrared spectroscopy (fNIRS) which records hemodynamics of the brain tissue. Sensitivity of fNIRS to the brain activity is however being affected by natural physiological brain hemodynamics (systemic interferences). Functional hemodynamic signal extraction from physiological interferences still remains a challenging task.

New method

This paper presents a novel effective algorithm for real-time physiological interference reduction and recovery of the evoked brain activity using a dual channel fNIRS system.

Results

Performance of the proposed algorithm was evaluated using both synthetic and semi-real fNIRS data using three different metrics of: 1) correlation coefficient (CC), relative mean square error (rMSE), percentage estimation error of peak amplitude (${\mathrm{E}}_{\mathrm{PA}}$).

Comparison with existing methods

The results were compared to those of ensemble empirical mode decomposition based recursive least squares (EEMD-RLS) method which has proved to have a better performance than other widely used algorithms such as block averaging, band-pass filtering and principal and/or independent component analysis. This study showed that the proposed method outperforms the EEMD-RLS method producing a smaller average rMSE and ${\mathrm{E}}_{\mathrm{PA}}$ and a larger average CC even in the cases of shorter signal lengths and smaller signal to noise ratio conditions.

Conclusions

The proposed method has no assumption on the amplitude, shape and duration of the hemodynamic response such as those needed in other previously reported methods. Moreover, it is computationally low cost and simple, and needs no parameter updating.

Introduction

Functional Near-Infrared Spectroscopy (fNIRS) is a relatively new noninvasive neuroimaging technique to discover hemodynamic variations within the cortex [1], [2], [3], [4]. It is an increasingly popular technology for brain function assessment that is due to its several advantages over other techniques (like EEG, fMRI, etc.) for being low-cost, portable, safe, having relatively high temporal resolution and a quick setup [5], [6], [7].

fNIRS is based on the fact that activation of a specific localized part of the brain leads to an increase in oxygen consumption in that region, which is accompanied by an increase in the total blood flow, regional blood volume and regional blood oxygenation due to neurovascular coupling [8], [9]. This leads to a change in the concentration of the local oxygenated haemoglobin (oxy-Hb) and deoxygenated haemoglobin (deoxy-Hb) [10]. Since oxy-Hb and deoxy-Hb have specific optical properties in the near-infrared light range (between 700 and 900 nm), the change in the concentration of these chromophores during neurovascular coupling can be detected noninvasively. In fNIRS, the near infrared light photons passing through the tissue are either scattered or absorbed. Photons which are not absorbed travel a banana shaped path and back to the surface of the scalp, the amount of which can be measured using a photodetector. Choosing appropriate wavelengths with regards the absorption coefficients of oxy-Hb and deoxy-Hb, it is possible to calculate variations in the concentrations of these chromophores using a modified Beer–Lambert law (MBLL) [11], [12].

fNIRS has the ability to measure the time variations in both oxy-Hb and deoxy-Hb noninvasively and without imposing a major limitation to the subject. In spite of this and other advantages, the presence of the systemic physiological interferences in the cortex tissue is one of the main problems in using fNIRS to study the brain function. The so called systemic physiological interferences are present not only in the brain tissue but also in all other superficial tissues over the scalp. If the distance between the light source and the photodetector is set to about 3 cm over the scalp, photons penetrate to a depth of about 1.5 cm within the cortex. On the other hand, if a photodetector placed about only 1 cm away from the source on the scalp, only hemodynamic changes of the superficial layers, which are only the physiologic interferences will be recorded [13], [14]. The so called physiological interferences include cardiac pulsation, respiration, blood pressure and Mayer wave fluctuations. These will result in a poor estimation of the evoked hemodynamic responses (EHRs) by fNIRS [15], [16]. Motion artifact and electrical noise are other interferences affecting fNIRS signal.

Various algorithms have been developed for hemodynamic response extraction from fNIRS signals in the past, which are either based on single, dual and/or multichannel fNIRS instruments. Block averaging (BA) and bandpass filtering (BPF) methods have been used to reduce the systemic artifacts from a single channel fNIRS recordings. In the BA method, actual hemodynamic responses are retrieved by time domain averaging of the N EHRs recorded during N uniform stimuli [17], [18]. The higher the number of recorded EHRs the better will be the results in the cost of more processing time.

Based on the assumption that the frequency contents of the physiological interferences are different than those of EHRs’, some fNIRS based BCI studies have used a band-pass filter to remove the physiological interferences [5], [19]. This method is not however very effective since there are partial overlap in the low frequency components of the EHRs and the physiological interferences (i.e. respiration and the Mayer wave) [20].

Both principal component analysis (PCA) and independent component analysis (ICA) are prevalent methods used for EHR extraction from fNIRS signals in multichannel recordings [21], [22], [23]. These methods are based on the assumptions that the spatial distribution of the systemic interferences and the EHRs is different. While unlike the localization of brain functional activity, systemic physiological interferences are generally global and their effects can be seen in all recorded channels [24].

The main problem with both PCA and ICA algorithms is that the noise intercrosses from noisy channels to other channels introducing negative impact on the final results. Several methods for physiological interferences reduction from the fNIRS signal using a dual channel fNIRS system were reported in the past [25], [26], [27], [28], [29], [30], [31], [32], [33]. Adaptive filtering approach is the most common method to reduce the systemic interferences from the fNIRS signals in dual-channel fNIRS systems [34], [35], [36], [37]. In this method, signal of the near channel is considered as the reference input while signal of the far channel is considered as the primary input. Different algorithms are used to find appropriate scaling coefficients for the reference signal in order to subtract the scaled reference signal from the primary input in order to remove interfering trends.

In this article, we present a novel algorithm to extract the brain hemodynamic responses from the fNIRS signals obtained by a dual channel fNIRS instrument (including near and far channels) which is superior to the other previously reported methods in the sense that it has no assumption on the amplitude, shape and duration of the hemodynamic responses, it is computationally low cost and simple, and needs no parameter updating. We employed the wavelet transform to decompose the near and far signals into approximation and detail sub-bands each with distinct frequency content. Next, with an ad-hoc and simple method the physiological interferences are eliminated from the signal of the far channel. In order to evaluate the performance of the proposed method, the results of the proposed method were compared to those produced by the recently reported method of EEMD-RLS [37]. For this comparison the three parameters of relative mean square error (rMSE), percentage estimation error of peak amplitude (${\mathrm{E}}_{\mathrm{PA}}$) and Pearson's correlation coefficient (CC) were used as quantitative criteria. Application of the proposed method to the synthetic data showed a significant improvement in recovery of the EHRs. The results showed that even in the case of a high level of noise and a short data length the proposed method is very effective. In continue a semi-real data was generated by adding simulated EHRs to the real resting state data. Application of the proposed method to this data also showed a remarkable improvement in recovery of the evoked hemodynamic response.

The rest of the paper is organized as follows. Section 2 introduces the proposed algorithm. Section 3 presents the results of applying the proposed algorithm to the synthetic and semi-real data. Finally Section 4 concludes the paper.