Mapping distributed brain function and networks with diffuse optical tomography

Abstract

Mapping of human brain function has revolutionized systems neuroscience. However, traditional functional neuroimaging by positron emission tomography or functional magnetic resonance imaging cannot be used when applications require portability, or are contraindicated because of ionizing radiation (positron emission tomography) or implanted metal (functional magnetic resonance imaging). Optical neuroimaging offers a non-invasive alternative that is radiation free and compatible with implanted metal and electronic devices (for example, pacemakers). However, optical imaging technology has heretofore lacked the combination of spatial resolution and wide field of view sufficient to map distributed brain functions. Here, we present a high-density diffuse optical tomography imaging array that can map higher-order, distributed brain function. The system was tested by imaging four hierarchical language tasks and multiple resting-state networks including the dorsal attention and default mode networks. Finally, we imaged brain function in patients with Parkinson's disease and implanted deep brain stimulators that preclude functional magnetic resonance imaging.

Figure 1. The HD-DOT system

a, Imaging cap structure, subject position and audiovisual stimuli for HD-DOT (subset of optical fibres shown for clarity). b, Interleaved square grids of detectors (blue) and sources (cyan, magenta, green, yellow, orange and red) provide a regular lattice for efficient source-detector encoding. Colours of sources represent spatially defined encoding regions (S1, D1 and so on, denote the numbering of sources/detectors). c, Discrete detection channels, using APDs, provide a wide dynamic range (DR) of >10⁷ and low crosstalk of <10⁻⁶ (diamonds, APD-1; circles, APD-2; squares, APD-3; triangles, APD-4; plus symbols, APD-5; crosses, APD-6). d, Each source LED is frequency encoded by region and wavelength. e, Time traces for four detectors demonstrate the varying light level detected throughout a single encoding frame. Within an encoding region, the source positions are temporally encoded (sequentially). f, The SD-pair light levels have a log-linear fall-off versus distance, characteristic of photon diffusion through biological tissue (blue, 750 nm sources; green, 850 nm sources). g, The temporal variance in all ~1,200 SD-pairs is used to evaluate and optimize data quality. SD-pairs with variance below 7.5% (red line) are retained for imaging. h,i, Cap fitting (~8 min) maximizes the optode coupling power across the cap (h) and the regional distribution of SD-pairs that pass the noise threshold (i, solid lines). See Methods and Supplementary Sections I and II for details.

Figure 2. Finite-element modelling of NIR light propagation in a subject

a, To accurately model light propagation in a subject, the head model incorporates the head shape, internal tissue structure from the subject’s anatomical MRI and optode locations. b, The FOV for each subject has been spatially registered and overlaid on a cortical surface view of the Montreal Neurological Institute (MNI, McGill University) atlas. Colour bar: number of subjects with sensitivity at a given location of cortex. See Methods and Supplementary Sections III and IV for details.

Figure 3. Evaluation of distributed brain function mapping by imaging hierarchical language processing

a-d, Differential responses throughout the FOV in response to language tasks including hearing (a), reading (b), imagined speaking (c) and generating words (d) were spatially normalized and group averaged. The HD-DOT data (ΔHbO contrast) shows excellent agreement with non-concurrently recorded fMRI (smoothed with a 13 mm FWHM Gaussian kernel) in the same subjects. The maximum t-value (across subjects) is displayed below each cortex (t-maps thresholded at 50% maximum). The overlap of binary masks (reflecting the 50% cutoff in the t-values) is displayed for each stimulus response type. See Methods and Supplementary Section IV for details.

Figure 4. Mapping the functional connections of distributed brain networks

Functional connectivity maps were generated with HD-DOT (and fMRI) with seeds (black dots, MNI coordinates) representing three sensory-motor networks (visual (Vis) (a), auditory (Aud) (b), motor (Mot) (c)) and three higher-order networks (dorsal attention (DAN) (d), fronto-parietal control (FPC) (e) and default mode (DMN) (f)). g, Correlation matrices for the bilateral set of seeds shown in a-f. h, To explore the transition between networks, nine seeds were placed on a path connecting a seed in the occipital cortex (seed 1 in the visual network) to a seed within area MT (seed 9 in the DAN). Black scale bar, 1 cm. i, Strength of spatial correlation between a given seed FC map and the full set of seed FC maps. Colouring of lines reflects putative networks (blue, seeds 1-3 in the visual network; grey, seeds 4-6 within the transition zone; red, seeds 7-9 in the DAN). The spatial correlation between FC maps is strong for seeds within a network. The transition from the DAN to visual network occurs within ~1 cm. j, FC maps of five of the nine seeds from h are shown for HD-DOT and fMRI. Anterior-posterior FC characteristic of the DAN is apparent starting in seed map 7 (white circle).

Figure 5. Measuring brain function in subjects contra-indicated for MRI

a, Imaging of Parkinson’s patients with DBSs. A sagittal X-ray and an axial slice of a computed tomogram show the location of the bilaterally embedded electrodes in the subthalamic nucleus (arrows). b, Contrast-to-noise map of the response to hearing words measured in three patients with Parkinson’s disease. The number below the cortex denotes the maximum t-value (ΔHbO contrast). c-e, Example functional connectivity maps generated at the same seeds as Fig. 4 (seed locations shown on cortex, inset).