Concurrent intrinsic optical imaging and fMRI at ultra-high field using magnetic field proof optical components

Intrinsic optical imaging (IOI) is a well established technique to quantify activation-related hemodynamical changes at the surface of the brain, which can be used to investigate the underlying processes of BOLD signal formation. To directly and quantitatively relate IOI and fMRI, simultaneous measurements with the two modalities are necessary. Here, a novel technical solution for a completely in-bore setup is presented, which uses only magnetic field proof components and thus allows concurrent recordings with a quality similar to that obtained in separate experiments. Measurements of the somatosensory cortex of rats with electrical forepaw stimulation were used to verify this approach. The high spatial and temporal resolution of the fMRI data, which is possible due to the high magnetic field of 14.1 T, the use of a point-spread function-based distortion correction and optimized additional anatomical images, allowed accurate colocalization of the images of the two modalities. Accord-ingly, detailed investigations of the temporal and spatial relationships between the hemodynamic parameters and the fMRI signal, which demonstrate the linear dependence of the BOLD effect on changes in the concentrations of oxygenated and deoxygenated hemoglobin, are possible. Comparisons between the signals emerging from arterial, venous and parenchymal areas are possible and show clearly distinct characteristics. The presented setup allows combining MRI measurements and optical recordings without serious losses in the data quality of either modality. While the proposed combination of fMRI and IOI can help to gain valuable insight into the generation of the BOLD effect, the setup can be easily modified to include different types of optical or MRI measurements.


| INTRODUCTION
In spite of being applied in thousands of neuroscientific studies, the BOLD (blood oxygenation level dependent) effect and the underlying neurovascular coupling are still a question of intense research. Better knowledge of the temporal dynamics and the spatial distribution of the signal change due to brain activation could help to improve the interpretation of many neuroscientific experiments. In addition, to find an imaging technique that can serve to better localize the true area of activation is one of the main goals in functional MRI (fMRI). A major challenge is the complex generation of the BOLD effect as a combination of several interacting contributions, mainly the change in cerebral blood volume (CBV), cerebral blood flow (CBF) and blood oxygenation, which have different temporal and spatial distributions. Disentangling the influences of these parameters and their effect on the contrast of the selected imaging sequence could help a great deal to better localize brain activation in time and space, and many approaches have been undertaken to measure these different parameters independently by using different MRI techniques.
An alternative way of measuring changes in blood oxygenation and volume independently, fast and with high spatial resolution and high sensitivity is intrinsic optical imaging (IOI), which observes the brain surface illuminated by light of several wavelengths. Due to the different optical properties of oxygenated (HbO) and deoxygenated hemoglobin (HbR), changes in oxygenation and CBV can be quantified separately. While this technique is restricted to observing the surface of the brains of animals, due to the invasive characteristics and the limited penetration depth of the light used, its high spatial and temporal resolution, its high sensitivity and its ability to quantitatively measure the two parameters separately makes it an adequate tool to investigate the hemodynamic processes caused by brain activation and their effect on the BOLD signal.
A large number of studies [1][2][3][4][5][6][7][8][9][10][11] have so far used IOI in neuroactivation experiments to observe the spatial and temporal dynamics of the hemodynamic changes that cause the fMRI signal change. However, due to the strong magnetic fields required for fMRI, most studies that aim to directly compare fMRI and IOI data acquire the two modalities in separate measurements.
While sequential measurements have shown excellent agreement between the two techniques in finding activation patterns for evoked fMRI, [12][13][14] for studies investigating resting state fMRI or the quantitative relation between the hemodynamic parameters and the MRI signal, BOLD and IOI measurements have to be performed simultaneously to obtain data that can be directly compared and related. Consequently, Kennerley et al. 15 presented a setup for concurrent MRI and IOI examinations and applied it successfully to several research questions. [16][17][18][19] The instrumentation in these experiments was based on an endoscope, using a large optical fiber bundle to guide the light in and out of the magnet.
Recently, a similar setup was used to make it possible to simultaneously acquire fMRI and fluorescence-based Ca 2+ imaging. 20 We present here a somewhat different approach for concurrent fMRI and IOI measurements where we attempt to transfer the optimized instrumentation of standard IOI systems into the magnet bore, close to the animal to be examined. By using special, non-magnetic and MRI proof optical components, including the camera itself, we aim to obtain IOI data with the same quality as would be possible outside the magnet. By furthermore conducting our experiments in an MR scanner with an ultra-high field strength of 14.1 T, we are able to maximize the sensitivity and resolution of both BOLD and IOI simultaneously.

| MRI
The experiments were performed on a 14.1 T magnet (Magnex Scientific, Oxford, UK), equipped with an RRI (Resonance Research Inc., Billerica, MA, USA) gradient, which is able to reach a gradient strength of 1 T/m within 189 μs. The total bore size is 26 cm, which is reduced to a usable bore of 12 cm inside the gradient. The system is interfaced to a Bruker (Bruker BioSpin, Ettlingen, Germany) Avance 3 console running ParaVision 6.0.1.

| Optical imaging system
To enable IOI inside the strong magnetic field of a 14.1 T MRI system without having to sacrifice image quality compared with experiments outside the scanner, the original setup proposed by Bonhoeffer and Grinvald 21 was modified for magnetic field compatibility. The limited space inside the magnet (12 cm bore size inside the gradient) further had to be taken into account in the construction of the setup.
The system is based on a magnetic field proof, board level version of a commercially available scientific CMOS camera (pco.edge 4.2 WAT, PCO, Kelheim, Germany), placed in a home-designed bronze housing to shield off camera-related noise from the MR measurement and to allow for water-cooling of the sensor. Two lens systems (Sonnar 1.8/135, Carl Zeiss, Oberkochen, Germany, and Nikkor 50 mm, 1:1.2, Nikon Corporation, Tokyo, Japan) were carefully disassembled, relieved of all metallic parts (aluminum housing and lens fixation rings) and reassembled using only components manufactured from PEEK (polyester ester ketone), taking care to leave the geometry unchanged. The lenses were then put together to form a tandem lens combination, yielding a magnification factor of 2.7.
The animal is placed below the lenses and positioned and held in place by a stereotactic holder. A prism positioned above the brain deflects the light into the lens/camera system.
For illumination, LEDs with four different dominant wavelengths were used: 465 nm (blue, Cree XP-E2 blue, Cree, Durham, NC, USA), 520 nm (green, Cree XP-E2 green), 595 nm (yellow, Cree XP-E2 amber), 625 nm (red, Osram LR W5AM, Osram Opto Semiconductors, Regensburg, Germany). Two LEDs of each color were controlled by homemade high-speed LED drivers positioned outside the scanner room and their light was transmitted into the magnet by two optical fibers per wavelength, which illuminate the brain from above through the prism. For MRI transmit and receive, a circular surface coil with a diameter of 22 mm was attached below the prism, just above the brain. A glass conus leading through the coil served as an optical connector between the prism and the brain surface. Except for the RF-coil, no magnetic or metallic parts are in proximity to the imaged brain region, to avoid susceptibility induced distortions of the EPI images.
The entire setup, which is small enough to fit inside the magnet, is illustrated in Figure 1.
The optical imaging acquisitions are controlled by home written software, which drives the recording of a specified number of images at each trial and sets the illumination and stimulus. The synchronization with the pulse sequence is done using an Arduino microcontroller (Arduino Leonardo; Arduino, Somerville, MA, USA), which accepts trigger pulses from the sequence and sends a signal to the main software. A second Arduino is used to rapidly switch the LEDs controlled by output signals from the camera to alter the illumination wavelength between recording consecutive images. A third Arduino serves to modify the parameters of the electric forepaw stimulation by setting pulse current, frequency or duration of the stimulus within the monitoring and stimulus software (LabChart, ADInstruments, Sydney, Australia) under the control of the main experimental software. All experiments shown here acquired 600 images per trial with a frequency of 20 Hz, starting 6 s before the start of stimulation.

| Animal preparation
All animal experiments were performed in agreement with German and European regulations and approved by the local authorities. Male Lister hooded rats with weights between 170 g and 230 g were anaesthetized by continuous inhalation of sevoflurane (0.5-2%) and one bolus of urethane (1.25 g/kg body weight (BW) i.p.), with additional doses (0.25 g/kg BW) given as necessary (maximum 2.0 g/kg BW). Throughout the surgery and the imaging experiment, the rectal temperature was measured and maintained at 37 C by means of a homeothermic blanket (Harvard Apparatus, Holliston, MA, USA), the breathing frequency was monitored by a pressure sensor placed below the animal, and heartbeat and spO 2 were measured using a magnetic field proof pulse oximeter (MouseOx, Starr Life Sciences, Oakmont, PA, USA). In most experiments, a F I G U R E 1 Drawing of the optical system with the rat inside the magnet (top) and a photograph of the final setup. subcutaneous injection of buprenorphine (0.05 mg/kg) or tramadol (25 mg/kg) was given for pain relief. In addition, lidocaine was applied locally at the site of the surgery to further improve analgesia.
The skin above the left somatosensory cortex was removed and the skull over an approximately 5 Â 8 mm 2 window was thinned to translucency using a dental drill under constant cooling with saline. All bleeding was stanched carefully and coagulated blood was removed. The sevoflurane anesthesia was discontinued and another dose of urethane was applied if necessary. The rat was transferred to the optical imaging setup and non-magnetic electrodes were inserted into the right forepaw for electrical stimulation. The brain window was placed below the prism and the MRI surface coil; the space between the skull and the glass conus (approximately 0.2-1.6 mm, depending on animal size and anatomy) was completely filled with transparent agar (0.15%) for optical coupling and to moisten the site of surgery. During the functional experiments, the spontaneously breathing animal was supplied with a mixture of room air and oxygen (70:30), keeping SpO 2 values above 95% during the entire experiment.

| Concurrent BOLD and IOI during forepaw stimulation
After carefully adjusting the distance between camera and prism to set the optical focus plane and performing the usual MRI parameter optimiza- To facilitate overlay of BOLD and optical images, two high-resolution anatomical gradient echo images with a resolution of 104 μm (FOV (26.6 Â 26.6) mm 2 , matrix size 256 2 ) were acquired. The first one was optimized for depiction of veins (T E = 4 ms, T R = 120 ms, flip angle = 25 , six slices, slice thickness = 0.5 mm, six averages), the second one used inflow to visualize arteries (T E = 1.69 ms, T R = 4.3 ms, flip angle = 20 , one slice, slice thickness = 1.5 mm, 100 averages). In addition, two runs of a calibration scan for a point-spread function-based distortion correction 22,23 were performed before and after the fMRI experiment to allow for accurate superposition of BOLD and optical images.
Electrical forepaw stimulation was applied via needle electrodes inserted into the palm of the right forepaw, using the PowerLab (ADInstruments) stimulus isolator. In all experiments, one of the three parameters determining the stimulation was changed in an interleaved manner between trials: either the amplitude (0.7 mA, 1.3 mA, 1.9 mA, 2.5 mA), duration (1 s, 2 s, 3 s, 5 s) or frequency (3 Hz, 6 Hz, 9 Hz, 12 Hz) of the stimulation pulses was varied in separate experiments, leaving the other parameters constant (2.5 mA, 3 s, 9 Hz). 30 trials were acquired for each parameter value.
In total, 11 animals were included in the study. In each of them, up to three experiments with different varied parameters were performed depending on animal state (four animals with one, six with two and one with three experiments), adding up to a total of 18 experiments (four with varying stimulus duration, eight with varying stimulus amplitude and six with varying stimulus frequency).

| Postprocessing
The optical images were binned from the original 2560 Â 2160 pixels to 640 Â 540 and normalized with the mean over the last five images of the series, which are considered to represent the rest condition. Image series from the 30 trials for the same stimulus were averaged. Hemoglobin concentrations were then calculated from each set of four consecutive images with different illumination wavelengths, using the differential pathlength approach, 24-26 which takes into account the different scattering characteristics and thus the different pathlengths through the tissue for light with the individual wavelengths. The required values for pathlengths and absorption were taken from the literature. 27 The results were a series of 150 images with a temporal resolution of 200 ms and a pixel size of 12 μm, showing changes in the concentration of HbO and HbR and, as sum of the two, the total hemoglobin (HbT) concentration, which was shown to be equivalent to the local CBV 15,19 as measured with MRI techniques.
For the BOLD measurements, the EPI images were distortion-corrected with help of the point-spread function-encoding scans. The resulting time series were run through SPM (SPM12, Functional Imaging Laboratory, Wellcome Trust Centre for Neuroimaging, London, UK 28 ) to determine activated regions for each value of the varied stimulus parameter, using a rat-specific hemodynamic response function. 29 In addition, time series were averaged over all 30 trials per parameter setting.
Using venous and arterial structures visible in the raw or processed optical images, the anatomical images with venous or arterial contrast, the distortion-corrected EPI images and the undistorted reference image from the distortion correction procedure, the size and position of the optical images were adjusted manually with respect to the fMRI data with the aid of home-written software. The temporal relations between the two measurements were fine-tuned by using the LabChart recordings to determine the mean delay between the stimulus trigger from the sequence and the actual start of the stimulation.

| RESULTS
Interactions between the two imaging modalities were modest. In spite of the shielding of the camera, a comparison of temporal SNRs in a phantom measurement with the same EPI protocol as used in the in vivo fMRI experiments showed an decrease by 7.5% after turning on the camera system, and an additional 5.5% while the camera is recording, adding up to a total 12.6% loss in tSNR compared with simple fMRI measurements without optical imaging. Improving the shielding of the camera and the cables might help to further decrease this loss. For the optical images, recording inside the magnetic field or during fMRI acquisition does not cause visible losses in image quality due to vibration or reductions in SNR, contrast or brightness compared with acquisition outside the magnet.
For all experiments, the large veins could clearly be identified in the anatomical and optical images, allowing for accurate colocalization of the two modalities ( Figure 2). In addition, the distortion correction applied to the fMRI data made sure that anatomical and fMRI images also con-  F I G U R E 3 Optical image (wavelength = 595 nm) with overlaid BOLD t-map, and maps for changes in the concentrations of HbO, HbR and CBV as determined from the optical imaging data.
18 measurements in 11 animals), as can be seen clearly in the individually normalized time courses in Figure 5B. There is also no post-stimulus undershoot visible in any data.
In Figure 6, the effects of varying stimulus parameters are shown. For varying duration ( Figure 6A), short stimuli up to 3 s yield a signal that continues increasing after end of the stimulus in the same way in all four measures for another 0.5-2 s. The amplitude of the signal change first increases with stimulus duration, before BOLD and HbR saturate after 3 s, while HbT and HbO slightly grow further, in agreement with previous observations. 30 Increasing stimulus frequency ( Figure 6B) causes increasing changes in activation parameters up to a frequency of 9 Hz, as is expected for rats under urethane anesthesia. 31 Further increasing the stimulus frequency to 12 Hz only slightly enhances the activation effect further. A variation of the activation intensity by changing the stimulation current yields a steady increase of all measured activation parameters.
The correlation between BOLD signal and increase in hemodynamic parameters, averaged over the entire activated region as detected by SPM, is presented in Figure 7, where the change in the hemodynamic parameters is plotted against the resulting BOLD signal change. All graphs  The distortion correction and the high-resolution anatomical images with venous and arterial contrast allow a very accurate colocalization of fMRI and optical images, which makes it possible to separate signals from different anatomical origin. Figure 8 shows the correlation between the BOLD signal change and the hemodynamic parameters for the different compartments, which were segmented manually based on the optical F I G U R E 8 Compartment-specific signals. Top: optical image with venous (blue) and arterial (red) structures marked from an experiment with varying stimulus frequency. The vessels marked with bold lines are used for the selection of the corresponding signals; everything else that is considered activated by SPM is combined as tissue signal. Left: BOLD and optical parameters from arterial, venous and tissue compartments. Right: corresponding correlations between BOLD and optical parameters for the different compartments.
images as well as the MR images with venous and arterial contrast. In the arteries, a small HbR variation and a large CBV change lead to a weak BOLD response, which is, however, expected to be mainly caused by an inflow effect due to the short repetition time in the fMRI measurements.
The venous compartment has strongest BOLD signal change evoked by a strongly increasing HbR with little CBV variation, while parenchyma signals vary moderately in spite of strong changes in HbO due to a limited variation in HbR. While qualitatively the observed relationships between the different compartments are expected, the concurrent acquisition of both modalities makes it possible to quantitatively correlate the different hemodynamic parameters and their effect on the fMRI signal, allowing us to closely investigate the generation of the BOLD effect.

| DISCUSSION
The formation of the BOLD effect has been investigated with a multitude of methods. As it is influenced by a variety of parameters, with blood flow, blood oxygenation and blood volume being only the most prominent ones, quantitative measurements of these parameters are a necessary prerequisite to investigate BOLD formation and to fuel and verify the simulations and models that are the basis of most predictions of fMRI investigations. Techniques of calibrated BOLD use MRI methods to obtain this data. However, the high variations of the fMRI activation signal between and even within experiments, especially for rodents under anesthesia, limit the quantitative accuracy of the outcome. Since simultaneous measurements of several hemodynamic parameters with MRI techniques are difficult, multimodal approaches may be able to help. IOI, as a proven tool with the ability to measure hemodynamic parameters quantitatively at high temporal and spatial resolutions, has been used extensively to investigate changes in blood volume and blood oxygenation during activation. Previous approaches for simultaneous fMRI and IOI have yielded highly interesting results and contributed significantly to understanding of the relation between BOLD and IOI: e.g., by showing the concordance of CBV as measured with MRI and HbT determined with IOI, 15,19 or by observing the formation of the negative BOLD effect with both techniques. 18 This study proposes an alternative approach for concurrent BOLD and IOI experiments by inserting all optical components into the MR magnet. This is possible due to a magnetic field proof version of a highly sensitive, commercially available, scientific CMOS camera, the standard version of which is frequently used in similar measurements outside magnetic fields. Using the also well established tandem objectives to ensure a defined and sharp focus plane opens the potential of obtaining the same optical quality as possible in standard, non-MRI IOI experiments. Since all commercially available lens systems contain metallic parts, which, even if non-magnetic, cause vibrations due to eddy currents induced while gradient switching, they had to be disassembled carefully and reassembled without compromising the geometrical consistency after replacing all metallic components. Electromagnetic shielding of the camera itself is also of vital importance to avoid degradation of the MRI signal. By placing the camera in a non-magnetic metallic box, the noise in the MR images could be suppressed to a large degree. The remaining SNR decline caused by the camera is acceptable, but should be removed by further improving the shielding of camera and cables. The only optical component outside the magnet is the illumination, for which LEDs with home-built, high-speed LED drivers, capable of switching LEDs within the 300 μs between two consecutive images, were used. Compared with the alternative approach of using a fiberoptic periscope to lead the light to a camera outside of the magnetic field, this integrated system is easier to handle and may avoid potential losses in SNR or resolution caused by the fiber bundle. In addition, it allows using the established tandem objective to guarantee a defined focus plane and suppress unwanted signals from other depths of the brain.
Since the goal was to colocalize IOI and fMRI data, the MRI acquisition parameters were carefully chosen to allow us to perfectly align those two data sets. The high SNR of the ultra-high field of our scanner made it possible to use a high spatial and temporal resolution. In particular, the slice thickness of 0.5 mm is small enough to ensure that the observed regions of BOLD and IOI, which predominantly collects light from only the superficial 0.4-0.8 mm of the cortex, coincide. The big differences in the BOLD distribution between the highest slice containing the surface veins and the one directly below it show that comparing signals from thicker slices to the optical imaging data may introduce strong discrepancies due to the mismatch in the volume from which the signals originate. The geometrical distortions, which are inherent in gradient-echo BOLD images and which would make an accurate overlay of the two image types impossible, are eliminated using a distortion correction approach based on a point-spread function prescan. High-resolution anatomical images with either venous or arterial contrast and the same orientation were acquired to further aid in spatial localization. All four image modalities-anatomical images with venous and arterial contrast, fMRI images and optical images-were shown to overlap accurately. This allows us to obtain and compare spatially selective signals of venous, arterial or tissue origin.
Finally, the need to synchronize camera and MRI recording made it necessary to write customized camera control software, which was also used to control all other aspects of the experiment. With three additional Arduino microcontrollers to control illumination, camera recording and stimulus, we were able to run our experiments with a duration of 1 h 40 min, automatically switching four different interleaved stimuli and acquiring more than 600 GB of optical data without user interaction.
In our experiments, we do not observe either an early deoxygenation effect, or a post-stimulus under-or overshoot. The latter may be due to the hyperoxic conditions during the measurements, as it has been shown that hyperoxia suppresses the undershoot. 7 This may also cause the lack of an initial dip in BOLD or an initial bump in the HbR data, although many studies with both IOI 6 and fMRI 32 fail to show these disputed effects of early deoxygenation in anaesthetized rats.
The quantitative comparison showed a linear dependence between the BOLD signal change and the hemodynamic parameters. This affirms the frequently used assumption of a β-exponent of 1 in the Davis model 33 of the BOLD effect for field strengths of and above 7 T. 34,35 Values of β at different fields have so far been mainly based on the results of Monte Carlo simulations 33,36 and measurements on humans applying hypercapnic or hyperoxic challenges, 37 while previous studies in animals that used contrast agents to modify the susceptibility of blood in a defined way even resulted in values significantly below unity. 38,39 Here, we had for the first time the possibility to directly measure the dependence of the BOLD signal on changes in the concentration of HbR during stimulation.
The high temporal resolution of both MRI and optical measurements allows a detailed comparison of the dynamics of the different signal components. The time lag between the HbO and HbR responses, which was observed previously in IOI, 40 was here shown to be accompanied by a delay in the BOLD signal increase. This difference in the temporal dynamics should also be visible in BOLD and CBV time courses measured with purely MRI-based techniques. 30 Using high-resolution fMRI, point-spread function-based distortion correction and optimized anatomical images that clearly show veins and arteries allows us to accurately colocalize fMRI and optical images and to isolate signal from selected tissue types. This makes it possible to separately investigate the contributions of venous, arterial and tissue structures to the BOLD effect and may help to verify the results of fMRI simulations. With the current setup, however, we suffer from insufficient visibility of the arteries in the optical images. Improving the angiography protocols for better visibility also of smaller vessels may be necessary for better identification of arterial structures.
The quantification accuracy of the optical imaging data currently suffers from the illumination wavelengths used and their different penetration depths. In particular, the red light carries signal contributions also from significantly deeper tissue layers than the lower wavelengths, introducing some error to the HbO and HbR values found. While we do not expect this to be a severe limitation, it would be easily resolved by improving the match in the penetration depths between the chosen wavelengths. Similarly, a small additional inaccuracy may be due to the spectral widths of the LEDs used around their dominant wavelengths. Adding bandpass filters in the light path after the LEDs can thus further improve the quantification.
Variations of the presented experiments are easily possible. Instead of BOLD measurements, the optical data can be enhanced by highresolution anatomical images, by CBV measurements using either contrast-agent-based methods or VASO (vascular space occupancy) or by arterial spin labeling (ASL) measurements of blood flow. Comparison of MRI signal distributions can be used to detect the sensitivity of different imaging sequences to badly localized venous origin. In addition, the optical observations can be related to different MRI-observable parameters such as diffusion or anatomic details. Furthermore, the MRI data may be used to enhance IOI experiments, which suffer from the small volume they observe due to the low penetration depth of the light used, by adding information from deeper regions of the cortex or the inner brain.
Similarly, the optical imaging setup can easily be modified for fluorescence imaging, for example using genetically encoded indicators for calcium imaging, or calcium or voltage sensitive dyes for observing different aspects of brain activity. Alternatively, other optical imaging techniques such as speckle flow imaging for measuring blood flow could be realized with only small modifications.

| CONCLUSION
Multimodal techniques are key in investigating the neural and hemodynamic processes during neuroactivation. The presented approach for simultaneous measurement of fMRI and IOI can contribute to improving our understanding of the spatial and temporal distribution of the BOLD signal and help to adjust and optimize fMRI protocols and postprocessing. The experiments shown here demonstrate the ability to quantitatively relate data acquired with the two techniques; modifications for obtaining different optical or MRI contrasts are easily possible and can further enhance the toolbox available for investigating the processes involved in neuronal activity and neurovascular coupling.