Extended hierarchical Bayesian diffuse optical tomography for removing scalp artifact

Brain space image reconstruction of functional near-infrared spectroscopy using a Bayesian adaptive fused sparse overlapping group lasso model

Significance

Functional near-infrared spectroscopy (fNIRS) is a noninvasive technology that uses low levels of nonionizing light in the range of red and near-infrared to record changes in the optical absorption and scattering of the underlying tissue that can be used to infer blood flow and oxygen changes during brain activity. The challenges and difficulties of reconstructing spatial images of hemoglobin changes from fNIRS data are mainly caused by the illposed nature of the optical inverse model.

Aim

We describe a Bayesian approach combining several lasso-based regularizations to apply anatomy-prior information to solving the inverse model.

Approach

We built a Bayesian hierarchical model to solve the Bayesian adaptive fused sparse overlapping group lasso (Ba-FSOGL) model. The method is evaluated and validated using simulation and experimental datasets.

Results

We apply this approach to the simulation and experimental datasets to reconstruct a known brain activity. The reconstructed images and statistical plots are shown.

Conclusion

We discuss the adaptation of this method to fNIRS data and demonstrate that this approach provides accurate image reconstruction with a low false-positive rate, through numerical simulations and application to experimental data collected during motor and sensory tasks.

Diffuse optical reconstructions of functional near infrared spectroscopy data using maximum entropy on the mean

Functional near-infrared spectroscopy (fNIRS) measures the hemoglobin concentration changes associated with neuronal activity. Diffuse optical tomography (DOT) consists of reconstructing the optical density changes measured from scalp channels to the oxy-/deoxy-hemoglobin concentration changes within the cortical regions. In the present study, we adapted a nonlinear source localization method developed and validated in the context of Electro- and Magneto-Encephalography (EEG/MEG): the Maximum Entropy on the Mean (MEM), to solve the inverse problem of DOT reconstruction. We first introduced depth weighting strategy within the MEM framework for DOT reconstruction to avoid biasing the reconstruction results of DOT towards superficial regions. We also proposed a new initialization of the MEM model improving the temporal accuracy of the original MEM framework. To evaluate MEM performance and compare with widely used depth weighted Minimum Norm Estimate (MNE) inverse solution, we applied a realistic simulation scheme which contained 4000 simulations generated by 250 different seeds at different locations and 4 spatial extents ranging from 3 to 40\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\text {cm}^2$$\end{document}cm2 along the cortical surface. Our results showed that overall MEM provided more accurate DOT reconstructions than MNE. Moreover, we found that MEM was remained particularly robust in low signal-to-noise ratio (SNR) conditions. The proposed method was further illustrated by comparing to functional Magnetic Resonance Imaging (fMRI) activation maps, on real data involving finger tapping tasks with two different montages. The results showed that MEM provided more accurate HbO and HbR reconstructions in spatial agreement with the main fMRI cluster, when compared to MNE.

Evaluation of a personalized functional near infra-red optical tomography workflow using maximum entropy on the mean

In the present study, we proposed and evaluated a workflow of personalized near infra-red optical tomography (NIROT) using functional near-infrared spectroscopy (fNIRS) for spatiotemporal imaging of cortical hemodynamic fluctuations. The proposed workflow from fNIRS data acquisition to local 3D reconstruction consists of: (a) the personalized optimal montage maximizing fNIRS channel sensitivity to a predefined targeted brain region; (b) the optimized fNIRS data acquisition involving installation of optodes and digitalization of their positions using a neuronavigation system; and (c) the 3D local reconstruction using maximum entropy on the mean (MEM) to accurately estimate the location and spatial extent of fNIRS hemodynamic fluctuations along the cortical surface. The workflow was evaluated on finger-tapping fNIRS data acquired from 10 healthy subjects for whom we estimated the reconstructed NIROT spatiotemporal images and compared with functional magnetic resonance imaging (fMRI) results from the same individuals. Using the fMRI activation maps as our reference, we quantitatively compared the performance of two NIROT approaches, the MEM framework and the conventional minimum norm estimation (MNE) method. Quantitative comparisons were performed at both single subject and group-level. Overall, our results suggested that MEM provided better spatial accuracy than MNE, while both methods offered similar temporal accuracy when reconstructing oxygenated (HbO) and deoxygenated hemoglobin (HbR) concentration changes evoked by finger-tapping. Our proposed complete workflow was made available in the brainstorm fNIRS processing plugin-NIRSTORM, thus providing the opportunity for other researchers to further apply it to other tasks and on larger populations.

Joint direct estimation of hemodynamic response function and activation level in brain functional high density diffuse optical tomography

High density diffuse optical tomography has become increasingly important to detect underlying neuronal activities. Conventional methods first estimate the time courses of the changes in the absorption coefficients for all the voxels, and then estimate the hemodynamic response function (HRF). Activation-level maps are extracted at last based on this HRF. However, the error propagation among the successive processes degrades and even misleads the final results. Besides, the computation burden is heavy. To address the above problems, a direct method is proposed in this paper to simultaneously estimate the HRF and the activation-level maps from the boundary fluxes. It is assumed that all the voxels in the same activated brain region share the same HRF but differ in the activation levels, and no prior information is imposed on the specific shape of the HRF. The dynamic simulation and phantom experiments demonstrate that the proposed method outperforms the conventional one in terms of the estimation accuracy and computation speed.

Resting-State Functional Connectivity Estimated With Hierarchical Bayesian Diffuse Optical Tomography

Resting-state functional connectivity (RSFC) has been generally assessed with functional magnetic resonance imaging (fMRI) thanks to its high spatial resolution. However, fMRI has several disadvantages such as high cost and low portability. In addition, fMRI may not be appropriate for people with metal or electronic implants in their bodies, with claustrophobia and who are pregnant. Diffuse optical tomography (DOT), a method of neuroimaging using functional near-infrared spectroscopy (fNIRS) to reconstruct three-dimensional brain activity images, offers a non-invasive alternative, because fNIRS as well as fMRI measures changes in deoxygenated hemoglobin concentrations and, in addition, fNIRS is free of above disadvantages. We recently proposed a hierarchical Bayesian (HB) DOT algorithm and verified its performance in terms of task-related brain responses. In this study, we attempted to evaluate the HB DOT in terms of estimating RSFC. In 20 healthy males (21-38 years old), 10 min of resting-state data was acquired with 3T MRI scanner or high-density NIRS on different days. The NIRS channels consisted of 96 long (29-mm) source-detector (SD) channels and 56 short (13-mm) SD channels, which covered bilateral frontal and parietal areas. There were one and two resting-state runs in the fMRI and fNIRS experiments, respectively. The reconstruction performances of our algorithm and the two currently prevailing algorithms for DOT were evaluated using fMRI signals as a reference. Compared with the currently prevailing algorithms, our HB algorithm showed better performances in both the similarity to fMRI data and inter-run reproducibility, in terms of estimating the RSFC.

Development of multi-directional functional near-infrared spectroscopy system for human neuroimaging studies

Multi-directional measurement using multi-directional light sources and multi-directional photodetectors drastically increases the amount of observation data without increasing the number of optical probes. In this study, we developed a novel multi-directional functional near-infrared spectroscopy (fNIRS) system for human neuroimaging studies. We tested our system by measuring the cortical hemodynamic changes of a single subject during a motor task and compared them with the same subject's functional magnetic resonance imaging (fMRI) data. We detected the direction-dependent fNIRS signals that originate from the cortical hemodynamic changes that are consistent with the fMRI data.

A Kalman-based tomographic scheme for directly reconstructing activation levels of brain function

In functional near-infrared spectroscopy (fNIRS), the conventional indirect approaches first separately recover the spatial distribution of the changes in the optical properties at every time point, and then extract the activation levels by a time-course analysis process at every site. In the tomographic implementation of fNIRS, i.e., diffuse optical tomography (DOT), these approaches not only suffer from the ill-posedness of the optical inversions and error propagation between the two successive steps, but also fail to achieve satisfactory temporal resolution due to the requirement for a complete data set. To cope with the above adversities of the indirect approaches, we propose herein a direct approach to tomographically reconstructing the activation levels by incorporating a Kalman scheme. Dynamic simulative and phantom experiments were conducted for the performance validation of the proposed approach, demonstrating its potentials to improve the calculated images and to relax the speed limitation of the instruments.

Sparsity-regularized approaches to directly reconstructing hemodynamic response in brain functional diffuse optical tomography

In brain functional diffuse optical tomography, conventional indirect approaches first separately reconstruct the spatial changes in the absorption coefficients at every time point and then calculate the spatial excited levels in terms of hemodynamic models. Direct approaches combine the two steps necessary in the indirect approaches and obtain the spatial excited levels directly. Although reconstruction quality has been improved by the direct approaches to some extent, they still lack sharp edges and suffer from low spatial resolution because of the ill-posedness of the inverse problems. In this paper, a priori sparsity is introduced to obtain the sparse solutions and further improve reconstruction quality. Simulation experiments are conducted to illustrate the expected performance improvements of the proposed approaches.

Overview of diffuse optical tomography and its clinical applications

Near-infrared diffuse optical tomography (DOT), one of the most sophisticated optical imaging techniques for observations through biological tissue, allows 3-D quantitative imaging of optical properties, which include functional and anatomical information. With DOT, it is expected to be possible to overcome the limitations of conventional near-infrared spectroscopy (NIRS) as well as offering the potential for diagnostic optical imaging. However, DOT has been under development for more than 30 years, and the difficulties in development are attributed to the fact that light is strongly scattered and that diffusive photons are used for the image reconstruction. The DOT algorithm is based on the techniques of inverse problems. The radiative transfer equation accurately describes photon propagation in biological tissue, while, because of its high computation load, the diffusion equation (DE) is often used as the forward model. However, the DE is invalid in low-scattering and/or highly absorbing regions and in the vicinity of light sources. The inverse problem is inherently ill-posed and highly undetermined. Here, we first summarize NIRS and then describe various approaches in the efforts to develop accurate and efficient DOT algorithms and present some examples of clinical applications. Finally, we discuss the future prospects of DOT.

Diffuse optical tomography using multi-directional sources and detectors

Diffuse optical tomography (DOT) is an advanced imaging method used to visualize the internal state of biological tissues as 3D images. However, current continuous-wave DOT requires high-density probe arrays for measurement (less than 15-mm interval) to gather enough information for 3D image reconstruction, which makes the experiment time-consuming. In this paper, we propose a novel DOT measurement system using multi-directional light sources and multi-directional photodetectors instead of high-density probe arrays. We evaluated this system's multi-directional DOT through computer simulation and a phantom experiment. From the results, we achieved DOT with less than 5-mm localization error up to a 15-mm depth with low-density probe arrays (30-mm interval), indicating that the multi-directional measurement approach allows DOT without requiring high-density measurement.

Time-frequency dynamics of the sum of intra- and extracerebral hemodynamic functional connectivity during resting-state and respiratory challenges assessed by multimodal functional near-infrared spectroscopy

Monitoring respiratory processes is important for evaluating neuroimaging data, given their influence on time-frequency dynamics of intra- and extracerebral hemodynamics. Here we investigated the time-frequency dynamics of the sum of intra- and extracerebral hemodynamic functional connectivity states during hypo- and hypercapnia by using three different respiratory challenge tasks (i.e., hyperventilation, breath-holding, and rebreathing) compared to resting-state. The sum of intra- and extracerebral hemodynamic responses were assessed using functional near-infrared spectroscopy (fNIRS) within two regions of interest (i.e., the dorsolateral and the medial prefrontal cortex). Time-frequency fNIRS analysis was performed based on wavelet transform coherence to quantify functional connectivity in terms of positive and negative phase-coupling within each region of interest. Physiological measures were assessed in the form of partial end-tidal carbon dioxide, heart rate, arterial tissue oxygen saturation, and respiration rate. We found that the three respiration challenges modulated time-frequency dynamics differently with respect to resting-state: 1) Hyperventilation and breath-holding exhibited inverse patterns of positive and negative phase-coupling. 2) In contrast, rebreathing had no significant effect. 3) Low-frequency oscillations contributed to a greater extent to time-frequency dynamics compared to high-frequency oscillations. The results highlight that there exist distinct differences in time-frequency dynamics of the sum of intra- and extracerebral functional connectivity not only between hypo- (hyperventilation) and hypercapnia but also between different states of hypercapnia (breath-holding versus rebreathing). This suggests that a multimodal assessment of intra-/extracerebral and systemic physiological changes during respiratory challenges compared to resting-state may have potential use in the differentiation between physiological and pathological respiratory behavior accompanied by the psycho-physiological state of a human.

Hierarchical Bayesian Model for Diffuse Optical Tomography of the Human Brain: Human Experimental Study

Diffuse optical tomography (DOT) is an emerging technology for improving the spatial resolution of conventional multi-channel near infrared spectroscopy (NIRS). The hemodynamics changes in two distinct anatomical layers, the scalp and the cortex, are known as the main contributor of NIRS measurement. Although any DOT algorithm has the ability to reconstruct scalp and cortical hemodynamics changes in their respective layers, no DOT algorithm has used a model characterizing the distinct nature of scalp and cortical hemodynamics changes to achieve accurate separation. Previously, we have proposed a hierarchical Bayesian model for DOT in which distinct prior distributions for the scalp and the cortical hemodynamics changes are assumed and then verified the reconstruction performance with a phantom experiment and a computer simulation of a real human head model (Shimokawa et al. 2013, Biomedical Optical Express). Here, we investigate the reconstruction accuracy of the proposed algorithm using human experimental data for the first time. We measured the brain activities of a single subject during a finger extension task with NIRS and fMRI. Our DOT reconstruction was compared with the fMRI localization results. Consequently, a remarkable consistency between fMRI and our DOT reconstruction was observed both in the spatial and temporal patterns. By extending the advantages of NIRS such as low running cost and portability with our DOT method, it might be possible to advance brain research in a real environment, which cannot be done with fMRI.

State-space models of impulse hemodynamic responses over motor, somatosensory, and visual cortices

THE PAPER PRESENTS STATE SPACE MODELS OF THE HEMODYNAMIC RESPONSE (HR) OF FNIRS TO AN IMPULSE STIMULUS IN THREE BRAIN REGIONS: motor cortex (MC), somatosensory cortex (SC), and visual cortex (VC). Nineteen healthy subjects were examined. For each cortex, three impulse HRs experimentally obtained were averaged. The averaged signal was converted to a state space equation by using the subspace method. The activation peak and the undershoot peak of the oxy-hemoglobin (HbO) in MC are noticeably higher than those in SC and VC. The time-to-peaks of the HbO in three brain regions are almost the same (about 6.76 76 ± 0.2 s). The time to undershoot peak in VC is the largest among three. The HbO decreases in the early stage (~0.46 s) in MC and VC, but it is not so in SC. These findings were well described with the developed state space equations. Another advantage of the proposed method is its easy applicability in generating the expected HR to arbitrary stimuli in an online (or real-time) imaging. Experimental results are demonstrated.