HIGH-DENSITY DIFFUSE OPTICAL TOMOGRAPHY: Imaging distributed function and networks in the human brain

Adam T. Eggebrecht and Joseph P. Culver

Mapping of human brain function has revolutionized our understanding of the brain. Brain imaging via positron emission tomography (PET) or functional magnetic resonance imaging (fMRI) have revealed the biological basis of human behavior: Brain systems that support everything we do—from sensing the visual world and generating language to daydreaming or sleeping—are accessible to investigation because of these techniques.^1-3 However, PET uses ionizing radiation, and this exposure risk is not permitted as an experimental procedure in children. And fMRI involves exposure to strong magnetic fields and induced electric currents, so is contraindicated in patients with implanted electronic devices (e.g., deep brain stimulators, cardiac pacemakers, and cochlear implants). Additionally, these modalities are not portable and thus are severely limited in clinically interesting cases like monitoring brain health in the intensive care unit or during surgery in the operating room.

Optical neuroimaging offers a portable, inexpensive, and noninvasive alternative that is radiation-free and compatible with implanted metal and electronic devices. Optical methods use near-infrared spectroscopy (NIRS), a safe technique (employed in pulse oximeters) that measures light absorption at multiple wavelengths to report blood volume and oxygenation.⁴ Living biological tissue is relatively transparent to NIR wavelengths of light and-somewhat surprisingly—the NIR light can diffuse multiple centimeters through an individual's head, including the skull. By placing NIR source/detectors a centimeter or more apart on the head, it is possible to record a time—varying signal that reflects changes in hemodynamics within the underlying tissue. While some of this signal is due to changes in oxygenation within the scalp and skull, a significant portion is driven by functional changes in hemodynamics within the brain.

Neuronal activity in the brain results in a cascade of processes that locally increases cerebral blood flow and alters the relative concentrations of oxygenated (HbO) and deoxygenated (HbR) blood. This altered balance of HbO and HbR provides the blood-oxygen-level-dependent (BOLD) signal measured with fMRI.⁵ Thus, the optical spectroscopy of blood by NIRS provides a window into brain function similar to that offered by fMRI. Traditional functional NIRS (fNIRS) imaging uses sparse arrangements of NIR photon source-detector measurements that have significantly lower spatial resolution than fMRI. Recent developments in high-density diffuse optical tomography (HD-DOT), which uses multiple overlapping NIRS measurements, have provided dramatically improved resolution and brain sensitivity.⁶

Despite these advances and because of limited field of view, HD-DOT has not yet successfully simultaneously imaged multiple distributed functional systems (e.g., responses in primary visual areas along with 'higher-order' cognitive responses such as language generation or control of attention).^7,8 Further, HD-DOT initially suffered from a lack of anatomical data registration. But all of that is changing.

Design for clinical priorities

Researchers in the Optical Radiology Lab at Washington University School of Medicine in St. Louis recently reported a HD-DOT imaging system that overcomes a number of technical challenges through integrative advances in optical instrumentation, fiber-optic cap design, optical data registration, and light modeling.⁹ In designing the HD-DOT system, we focused on solving the challenges most crucial to recovering maps of brain function: resolution, brain specificity, an extended field of view (FOV), and anatomical specificity/co-registration of optical data. The system covers 2/3 of the head, using 96 source and 92 detector positions to provide >1200 usable NIRS measurements (see Fig. 1). A discrete avalanche photodiode (APD) detector channel design provides a signal-to-noise ratio (SNR) that is adequate (greater than 100:1) over a dynamic range of four orders of magnitude in light level. Due to the tight packing, each source and detector supports multiple measurements, thus providing significant challenges in illumination encoding and decoding. To achieve a full-field frame rate of 10 Hz with crosstalk below 10^-6, we tagged source light with a mixture of spatial, frequency, and temporal encoding.

Maintaining stable fiber-optic connections with the head poses significant challenges to cap design-the objectives being optimal light coupling, efficient cap fitting (< 15 min.), and subject comfort. The design strategy centers around a cap structure that decouples lateral torque of the fibers (parallel to the head surface) from longitudinal flexibility (perpendicular to the head) to enable the cap to both conform to the head surface and maintain good optic-scalp coupling. To achieve this, the 188 fiber bundles are suspended above the subject's chair to manage fiber organization and to provide distributed support of the fiber weight around the head. Optical fiber tips extend through the interior of the cap ~3 mm to comb through the hair and couple directly to the scalp.

Optical data registration with the anatomy of the subject can be obtained with the use of an individual subject's MRI-based anatomy, or via an atlas head model that is mathematically registered to the subject's head shape (see Fig. 2). The use of an atlas model has been validated by multiple labs and facilitates the use of DOT in settings where MRI-based anatomy may not be available; for example, within the neonatal ICU or if the subject has implanted devices that preclude MRI.^10-12The sensitivity of each NIRS measurement pair is then calculated in the head model.¹³ Finally, the optical data can be reconstructed to generate co-registered images of brain activity, such as in response to hearing words.

FIGURE 2. For data modeling and registration, DOT methods can use (a, b) either an individual subject's MRI-based anatomy, or else an atlas head model that is mathematically registered to the person's head shape. (c) The light modeling panel shows sensitivity of a source (red) and detector (blue) with the color map and contours (respectively). (d) The far right image shows DOT-detected brain activity in auditory cortex in response to hearing words.

These technology advances have enabled the HD-DOT system to perform functional brain mapping akin to PET and fMRI within the outer centimeter of brain tissue. For example, when subjects are asked to read words, imagine speaking words, or to generate novel words, the brain lights up in the HD-DOT system in the same places that it does when the same subjects are lying in an fMRI scanner (see Fig. 3). These blushes of brain activity highlight the visual cortex, motor cortex, and left-lateralized Broca's area, respectively.

FIGURE 3. DOT maps of language processing validated against fMRI.

Stimulus-based investigations of the brain, as with these language tasks, have provided powerful insight into many aspects of brain function. However, a complementary view of brain function considers that most brain activity is continuously ongoing, involving information processing for interpreting, responding to, and predicting the world around us. Measuring and analyzing this intrinsic activity has led to another revolution in understanding the functional structure and organization of the brain.² Importantly, the methods of analysis of this intrinsic activity, called functional connectivity, can be applied to those who are unable to respond to tasks, such as neonates or unconscious subjects.

To further validate data quality of the HD-DOT system against fMRI image quality, we generated maps of key so-called resting-state networks using functional connectivity analysis (see Fig. 4). The maps demonstrate clearly separable regions associated with visual and motor processing that are in agreement between HD-DOT and fMRI. These sensory and motor networks are generally composed of regions that are duplicated across hemispheres, but are relatively simple within a hemisphere. This is contrasted by resting state networks that perform cognitive functions such as control over processing and attention (in the frontoparietal network), or self-reflection and daydreaming (in the default mode network). These networks are composed of spatially distributed regions within and across the hemispheres of the brain (multiple yellow regions in Fig. 4).

FIGURE 4. As part of an effort to validate data quality of HD-DOT against fMRI, we generated maps of 'resting-state' activity in the relatively simple visual and motor processing networks (left) and the more complex networks that perform cognitive functions such as control over processing and self-reflection (right). — FIGURE 4. As part of an effort to validate data quality of HD-DOT against fMRI, we generated maps of "resting-state" activity in the relatively simple visual and motor processing networks (left) and the more complex networks that perform cognitive functions such as control over processing and self-reflection (right).

Appropriate application

HD-DOT, unlike fMRI, is restricted to imaging superficial cortex, thus limiting direct access to brain function in deep sub-cortical brain structures (e.g., striatum and thalamus) or deep cortical structures (e.g., insula and operculum). This limitation is potentially a problem in mapping functional connectivity throughout the brain, as HD-DOT cannot access all areas in some networks. However, every known resting state network has regions in the superficial cortex. Therefore, although deep structures are out of range, a lesion within one area of the brain often will have network-wide effects that can be detected via disruptions in functional connectivity measured within the FOV accessible to HD-DOT. Thus, while constrained to superficial cortex, HD-DOT can potentially monitor the brain's functional network dynamics in at least portions of all of the major described networks.

A recent special issue on fNIRS by the journal NeuroImage captured a snapshot of the great advancements made during the years since the first seminal papers on fNIRS in 1993; the breadth and variety of studies summarized in the issue point towards many extensions to the results of the 1993 paper.⁵ While the present cap is more comfortable than previous smaller caps owing to balanced weight management, wearability could be greatly advanced over the present system with the development of lighter-weight caps, smaller fiber optics, or even placing electro-optic elements on the cap.¹⁴ And the image quality demonstrated here with hemoglobin contrasts could potentially be extended to other optical contrasts;¹⁵ for example, recent fluorescent measurements in humans¹⁶ suggest the possibility of HD-DOT using molecular imaging contrasts.¹⁷ Similarly, recent transcranial measurements of light coherence during functional activations point toward tomography of cerebral blood flow in humans.^18,19

FIGURE 5. DOT enables (a) within-room social interactions, (b) neuroimaging within the operating room, and (c) bedside neuroimaging of neonates.

As the image quality of HD-DOT improves, the attractive features of optical techniques can be further leveraged. Besides greater safety and flexibility, HD-DOT also offers a more ecologically natural scanning environment: Whereas MRI scanner noise can exceed 120 dB, HD-DOT is effectively silent (~10 dB), which permits more nuanced language studies that use stimuli delivered with the volume of normal conversation. For instance, HD-DOT can image with subjects sitting in a comfortable chair, thus making possible studies on direct within-room natural social interactions potentially useful in studies of disorders such as autism (see Fig. 5).²⁰ Further, the potential portability of HD-DOT offers opportunities for bedside monitoring in critical care environments (e.g., the operating room or in neonatal intensive care).

All of this points towards noninvasive optical neuroimaging as a robust, powerful, and practical tool.

REFERENCES

1. S. E. Petersen, P. T. Fox, M. I. Posner, M. Mintun, and M. E. Raichle, Nature, 331, 6157, 585–589 (1988).

2. M. E. Raichle, Trends Cognitive Sci., 14, 4, 180–190 (2010).

3. M. Corbetta and G. L. Shulman, Nat. Rev. Neurosci., 3, 3, 201–215 (2002).

4. F. F. Jobsis, Science, 198, 4323, 1264–1267 (1977).

5. D. A. Boas, C. E. Elwell, M. Ferrari, and G. Taga, NeuroImage, 85, 1, 1–5 (2014).

6. B. W. Zeff, B. R. White, H. Dehghani, B. L. Schlaggar, and J. P. Culver, Proc. Nat. Acad. Sci. USA, 104, 29, 12169–12174 (2007).

7. S. P. Koch et al., Frontiers Neuroenerg., 2, 1–12 (2010).

8. A. T. Eggebrecht et al., NeuroImage, 61, 4, 1120–1128 (2012).

9. A. T. Eggebrecht et al., Nat. Photon., 8, 448–454 (2014). (PMCID in progress).

10. A. Custo et al., NeuroImage, 49, 1, 561–567 (2010).

11. R. J. Cooper et al., NeuroImage, 62, 1999–2006 (2012).

12. S. L. Ferradal, A. T. Eggebrecht, M. Hassanpour, A. Z. Snyder, and J. P. Culver, NeuroImage, 85, 117–126 (2014).

13. H. Dehghani et al., Comm. Numer. Meth. Eng., 25, 6, 711–732 (2008).

14. S. K. Piper et al., NeuroImage, 85, 1, 64–71 (2014).

15. B. Hallacoglu et al., J. Biomed. Opt., 17, 8, 081406–081401 (2012).

16. C. Habermehl, C. H. Schmitz, and J. Steinbrink, Opt. Exp., 19, 19, 18636–18644 (2011).

17. V. Ntziachristos, A. H. Hielscher, A. G. Yodh, and B. Chance, IEEE Transact. Med. Imaging, 20, 6, 470–478 (2001).

18. A. Liebert et al., NeuroImage, 24, 2, 426–435 (2005).

19. L. Meng et al., Can. J. Anaesth., 59, 4, 357–365 (2012).

20. S. Lloyd-Fox et al., Proc. R. Soc. B, 280, 1758 (2013).

ADAM T. EGGEBRECHT, Ph.D., is a postdoctoral research scholar and JOSEPH P. CULVER is associate professor in the Optical Radiology Lab at Washington University School of Medicine (St. Louis, MO); e-mail: [email protected]; http://orl.wustl.edu/.