Keywords

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

5.1 Introduction

The ability to visualize and monitor dynamic blood flow changes in tissue is of great importance for a wide range of applications and diseases, especially during surgical procedures. Vascular injury or irreversible tissue damage can result if a surgeon inadvertently occludes a vessel and does not repair the blockage quickly. This is particularly important when the tissue of interest is the brain, since prolonged ischemia can result in postoperative functional deficits. Thus, monitoring cerebral blood flow (CBF) during neurosurgery provides important physiological information and can help improve surgical outcomes for a wide variety of procedures. In order to minimize the amount of functional loss incurred by patients undergoing brain surgery, surgeons must be able to assess cortical function relative to the pathology to maintain a balance between sufficient resection of pathological tissue and preservation of motor, language, and sensory function. For procedures that require the removal of diseased brain tissue such as tumor resection or epilepsy, monitoring CBF can be used to localize areas of eloquent brain tissue and to assess postsurgical tissue viability. Monitoring CBF is particularly important during cerebrovascular procedures such as aneurysm clipping or vessel bypass, where the surgeon is operating directly on the blood vessels. The surgeon must be able to assess whether blood flow has returned to pre-surgical baseline levels in parent and branching vessels and identify residual stenosis or occlusion intraoperatively to reduce the risk of ischemic stroke and irreversible brain damage. Currently, surgeons rely on qualitative visual inspection under standard visible light, or use one of the clinically available intraoperative vascular monitoring technologies. In most cases, vessel occlusions are not visible to the naked eye, making visual assessment an unreliable monitoring solution. Although there are multiple intraoperative vascular monitoring techniques available, none of them are dependable under all circumstances and improvement is still needed to address the existing shortcomings.

The most widely used intraoperative vascular monitoring tools include the gold standard intraoperative digital subtraction angiography (DSA) [1–4], electrophysiological monitoring [5–7], neuroendoscopy [8–10], microvascular Doppler sonography [11–13], and indocyanine green angiography (ICGA) [14–17]. Recently, Gruber et al. [18] evaluated these techniques for use during aneurysm clipping procedures, concluding that the techniques were “complementary rather than competitive in nature” and all had disadvantages when used as a stand-alone method. Relative to the other available techniques, near-infrared ICGA has emerged as a popular and important intraoperative tool only recently after integration into surgical microscope platforms from multiple manufacturers [19–21]. This provides a simple, real-time method for examining intravascular fluorescent activity in the operating field and allows the surgeon to visualize vascular blockages or stenosis. However, ICGA does not directly provide quantitative flow information and requires the injection of a fluorescent dye. This limits the use of the technique to only a few instances during a given surgery and requires a delay between repeat uses due to the presence of residual dye in the vasculature, meaning that perfusion cannot be continuously monitored. Also, patients who have an iodine allergy or have significant liver disease may be ineligible for safe use of the contrast agent [22]. Thus, the ability to obtain similar information provided by ICG fluorescence without the need for a contrast agent could be a major advantage, both for surgeons who will be able to continuously visualize perfusion during the procedure and for patients who are not a candidate for the ICG injection.

Optical techniques for monitoring flow in the clinical environment are based on dynamic light scattering and utilize either the Doppler Effect or time-varying laser speckle [23]. Laser Doppler techniques provide quantitative flow information, but are typically limited to single point measurements that require mechanical scanning to provide spatial information [24–26]. Full-field laser Doppler instruments have recently been developed with speeds up to 12 frames per second [27, 28] and Raabe et al. [29] used a system to image CBF intraoperatively, demonstrating functional activation areas consistent with fMRI and electrostimulation. Laser speckle imaging is based on the same phenomenon and provides a significant advantage in temporal resolution compared to laser Doppler systems. Because rapid processing techniques are available for computing and displaying the speckle contrast images [30], the temporal resolution of a laser speckle contrast imaging (LSCI) system is determined by the frame rate of the camera and may be >200 Hz for high-speed cameras available today. This method provides a relatively simple technique for visualizing detailed spatiotemporal dynamics of blood flow changes in real time with high spatial resolution. LSCI has been used for a large number of blood flow imaging applications in tissue, including the retina, skin, and brain. These tissues are particularly well suited for LSCI because the vasculature of interest is generally superficial. LSCI is unable to sense blood flow deeper than a few hundred microns because of the illumination and detection geometry, meaning that visual access to the vasculature is required. Although shallow penetration depth is a limitation of the technique, it makes LSCI ideal for use during surgery when the tissue is already exposed. LSCI has been used intraoperatively during neurosurgery with promising results, both with a commercially available system [31] and with a microscope-integrated design [32, 33]. This chapter reviews the physics behind LSCI and describes its utility as an intraoperative tool for monitoring blood flow during neurosurgical applications.

5.2 Laser Speckle Imaging Physics

Laser speckle is a random interference pattern produced when coherent light scatters from a random medium. The laser light scatters from different positions within the medium and travels slightly different path lengths, resulting in constructive and destructive interference. When the backscattered light is imaged onto a camera sensor, the interference produces a 2D randomly varying intensity pattern known as speckle. If the random medium is made up of individual moving scatterers, such as red blood cells within the vasculature of heterogeneous tissue, the speckle pattern fluctuates in time as a result of phase shifts in the backscattered light. By imaging the time-varying speckle pattern onto a camera with an exposure time (5–10 ms) greater than the time scale of the speckle intensity fluctuations (typically <1 ms for biological tissues), the camera integrates the temporal intensity fluctuations, resulting in blurring of the speckle pattern. Thus, areas of higher motion have more rapid intensity fluctuations and therefore more blurring of the speckles during the camera exposure. Since the motion of the scattering particles is encoded in the dynamics of the speckle pattern, a measure of blood flow can be obtained by quantifying the spatial blurring of the speckle pattern. This is accomplished by calculating the local speckle contrast, K, defined as the ratio of the standard deviation, \( {\sigma_{\rm{s}}} \), to the mean intensity of pixel values, \( \langle I\rangle \), within small regions of the acquired image [34],

$$ K(T) = \frac{{{\sigma_{\rm{s}}}(T)}}{{\langle I\rangle }}, $$
(5.1)

where T is the exposure time of the camera. Thus, the speckle contrast is a measure of the local spatial contrast in the speckle pattern. The speckle contrast image is a spatially resolved map of the local speckle contrast and is calculated from the raw speckle image by computing this ratio at each pixel from the surrounding N × N pixel region (typically N = 7). Theoretically, the speckle contrast ranges between 0 and 1 when the speckle pattern is sampled properly [35], where K = 1 indicates no blurring and therefore no motion and K = 0 means that the scatterers are moving fast enough to average out all of the speckles.

A typical example of a raw speckle image and the computed speckle contrast are shown in Fig. 5.1, which was taken from a rat cerebral cortex under normal conditions. The raw speckle image illustrates the grainy appearance of the speckle pattern. The speckle contrast image, computed directly from the raw speckle image using (5.1), represents a 2D map of motion occurring in the tissue, which is primarily due to blood flow. Areas of higher baseline flow, such as large vessels, have lower K values and appear darker in the speckle contrast image. Although speckle contrast values are indicative of the magnitude of motion in the sample, they are not linearly proportional to speed or flow. The exact quantitative relationship between speckle contrast and underlying flow is a complex function that is not completely understood for biological tissue and remains an active area of research [36].

Fig. 5.1
figure 00051

(a) Example of a raw speckle image taken from a rat cerebral cortex with a 3 mm field of view, which shows a grainy pattern with some areas of spatial blurring visible. (b) The corresponding speckle contrast image is calculated using a 7 × 7 pixel sliding window, which shows the blood vessels on the surface of the brain with high spatial resolution. Although the raw image appears to contain little information, the speckle contrast image reveals a tremendous amount of information about the motion of the scattering particles in the sample

Full size image

The theory of correlation functions used in dynamic light scattering theory can be used to calculate the speckle correlation time \( {\tau_{\rm{c}}} \) from the speckle contrast values, which can then be related to the underlying flow or speed. The temporal fluctuations of speckles can be quantified using the electric field autocorrelation function \( {g_1}(\tau ) \). Because \( {g_1}(\tau ) \) is difficult to measure, the intensity autocorrelation function \( {g_2}(\tau ) \) is recorded and can be related to \( {g_1}(\tau ) \) using the Siegert relation [37]

$$ {g_2}(\tau ) = 1 + \beta |{g_1}(\tau ){|^2}, $$
(5.2)

where \( \beta \) is a normalization term that accounts for speckle averaging due to mismatch between speckle size and pixel size as well as polarization and coherence effects. The speckle correlation time \( {\tau_{\rm{c}}} \) is the characteristic decay time of the speckle autocorrelation function. The original relationship between K(T) and \( {\tau_{\rm{c}}} \) was first proposed by Fercher and Briers in 1981 [38] and is given by

$$ K(T,{\tau_{\rm{c}}}) = {\left( {\frac{{1 - {{\text{e}}^{{ - 2x}}}}}{{2x}}} \right)^{ {1/2}}}, $$
(5.3)

where \( x = T/{\tau_{\rm{c}}} \). This expression has been widely used in the literature since its original introduction. Recently, a more accurate expression has been proposed by Bandyopadhyay et al. [39] to account for speckle averaging effects and is given by

$$ K(T,{\tau_{\rm{c}}}) = {\left( {\beta \frac{{{{\text{e}}^{{ - 2x}}} - 1 + 2x}}{{2{x^2}}}} \right)^{{1/2}}}. $$
(5.4)

Using simplifying assumptions, the speckle correlation time \( {\tau_{\rm{c}}} \) is assumed to be inversely proportional to blood flow, meaning that a smaller \( {\tau_{\rm{c}}} \) corresponds to a faster moving particles (red blood cells) in the tissue [23, 40]. Thus, the measured speckle contrast values are converted to correlation time using either (5.3) or (5.4) and relative blood flow changes can be calculated by computing the change in \( {\tau_{\rm{c}}} \) from a baseline value [41].

5.3 Instrumentation

One of the reasons LSCI has become widely adopted is the relative simplicity of the instrumentation required to image blood flow. The hardware for LSCI consists of a coherent light source such as a laser diode for illuminating the tissue, a camera for detection of the backscattered light, and imaging optics to focus the light onto the camera sensor. A traditional laboratory LSCI setup is shown in Fig. 5.2a. The laser beam is expanded and adjusted to illuminate the tissue area of interest, which may vary from a few millimeters in a small animal model to several centimeters in a clinical setting. The laser beam may be angled towards the tissue surface or have a near normal incidence and the wavelength of the laser is typically in the red to near infrared region to minimize effects from hemoglobin absorption. The camera specifications for LSCI vary widely in the literature, but generally a standard CCD or CMOS camera can be used. Because high light levels reach the camera, high dynamic range cooled cameras are usually not required for LSCI and inexpensive 8-bit cameras can provide excellent images of blood flow [42, 43].

Fig. 5.2
figure 00052

(a) Traditional laboratory LSCI setup consisting of a laser diode, imaging optics, and a camera. (b) Intraoperative LSCI setup, where all components of the traditional setup are integrated into the Zeiss OPMI Pentero neurosurgical microscope (adapted from the Zeiss Pentero Manual version 9.3)

Full size image

To be clinically useful to surgeons, the LSCI technology must be integrated into the existing surgical platform. Although Hecht et al. [31] demonstrated intraoperative use with a standalone commercial LSCI instrument, the surgery must be interrupted to position the instrument over the surgical field of view. Thus, for use during neurosurgery, we integrated the LSCI hardware into the Zeiss Pentero neurosurgical microscope, which was convenient to use and allowed visualization of the tissue perfusion with minimal additional setup time [32, 33]. A schematic of the adapted microscope is shown in Fig. 5.2b. To illuminate the brain, the laser diode (\( \lambda = {66}0\;{\text{nm}} \), P = 130 mW) was incorporated into a microscope add-on laser adapter, which was attached to the bottom of the microscope head. A wavelength in the visible range was chosen due to the visible wavelength pass filters (400–700 nm) built-into the microscope. The laser beam travels linearly through the adapter until it encounters a curved mirror, which directs the beam towards the cortex as shown in the side view of Fig. 5.2b. The steering toggle on the laser adapter controls the angle of the mirror, which allows positioning of the laser beam on the tissue area of interest. The laser power was adjusted using a laser diode controller and was measured to be 28 mW/cm2 for typical settings used, which is far below the ANSI standard of 200 mW/cm2 for maximum permissible exposure to a visible laser beam [44]. An 8-bit camera (Basler 602f) was connected to one of the side viewing ports on the microscope using a c-mount camera adapter. This enabled the use of imaging optics built into the microscope to focus the camera and zoom into the area of interest on the tissue surface. The addition of the laser and camera did not interfere with the normal sterile draping of the microscope, meaning that the surgeon could use the microscope as desired during the operation. When the surgeon was ready to perform LSCI imaging, the microscope’s built-in illumination was turned off and the laser was turned on for imaging. During the image acquisition, the patient’s electrocardiogram (ECG) signal was also recorded from an existing anesthesia monitoring system in the operating room and was used during post-processing and analysis to help remove pulsatile motion artifacts. Integration of the equipment into an existing neurosurgical microscope allows real-time blood flow imaging with minimal interference to the procedure, which is an important consideration for intraoperative imaging applications.

5.4 Simultaneous Imaging of ICG Fluorescence and LSCI

As discussed previously, ICG fluorescence angiography has been widely examined in the neurosurgical community and has been integrated into the neurosurgical microscopes from multiple manufacturers. Thus, it is of great interest to compare LSCI and ICG fluorescence directly to examine the information obtained from both methods. In a rat model, we performed LSCI and ICG fluorescence imaging simultaneously, using two cameras and a 785 nm laser both as the excitation source for the ICG dye and the coherent source for LSCI. A dichroic beam splitter was used to direct the backscattered excitation light (785 nm) towards the LSCI camera and the ICG fluorescence emission light (820–835 nm) towards the ICG camera. The cameras were adjusted so that both imaged the same area of interest through an objective lens. Images were captured from a rat cerebral cortex after craniotomy and a change in physiology was introduced by occluding a vessel within the field of view. First, a comparison is illustrated for baseline normal physiological flow. In Fig. 5.3, baseline LSCI images are overlaid to show relative blood flow differences throughout the field of view as indicated by LSCI in (a) and ICG in (b). The LSCI overlay has been thresholded to show high flow areas where the flow rate is three times greater than the surrounding parenchyma tissue. The correlation time image was used to determine areas of high flow relative to the parenchyma tissue. The overlay identifies the largest vessels in the field of view, with dark red indicating fastest flow and yellow indicating slower flow. The ICG fluorescence can be quantified by the speed of the dye wash-in, measured by the time taken for the dye to reach maximum fluorescence (rise time). Thus, the ICG overlay has been thresholded to show the wash-in times of the ICG dye, with red indicating earlier wash-in (shorter rise time) and blue indicating later wash-in (longer rise time). This image allows separation of the arteries and veins in the field of view, which is important information during a surgical procedure [19]. By watching the wash-in of the dye, the direction of flow can also be determined, which is one major advantage of ICG fluorescence. This figure demonstrates that both methods are capable of imaging the vasculature within the field of view with similar spatial resolution and that both techniques provide different information regarding blood flow in the field of view.

Fig. 5.3
figure 00053

Relative blood flow overlaid onto a baseline LSCI image from a rat cerebral cortex acquired during simultaneous imaging of LSCI and ICGA. (a) LSCI relative blood flow overlay, where red indicates faster flow and yellow indicates slower flow relative to surrounding parenchyma tissue. (b) ICG dye rise time overlay, where red indicates an earlier wash-in and blue indicates a later wash-in, allowing separation of arteries and veins. Scale bar represents 0.5 mm

Full size image

Images were also captured during a localized vessel occlusion induced using a rose-bengal photothrombotic clot model. The rose-bengal was injected after baseline images were acquired and a 532 nm laser was focused through the objective to activate the dye and induce a local clot on a vessel of interest in the field of view. An example from a second animal in the study is shown in Fig. 5.4, where LSCI and ICG fluorescence images are shown before and after the clot is formed. Although the images appear similar, the two techniques are providing different information. LSCI represents an overall blood flow map of the area, with darker areas representing higher flow and lighter areas indicating slower flow. The ICG images show the quantity of the fluorescent dye in the vessels, with larger vessels resulting in a brighter signal due to increased amount of dye present and darker areas indicating weaker fluorescent signal. In the pre-occlusion images, the vasculature is clearly visualized by both techniques. After the clot is formed, the upper branching vessel indicated by the arrow disappears from the field of view in the LSCI image, indicating that blood flow has reduced in the branched vessel to levels comparable to the surrounding parenchyma tissue. In the ICG image, the signal from the branched vessel is much dimmer, which results because the dye is being blocked from getting to that region. This demonstrates that both methods were able to identify the occluded vessel and also highlights the differences between the two techniques. While LSCI is truly measuring the motion within the tissue sample, ICG fluorescence allows visualization of blood volume rather than blood flow. This is clear since part of the branched vessel is still visible in the ICG images, which is due to dye becoming trapped inside the formed clot. Although it is still possible to identify the occlusion in this case, this may make ICG images more difficult for surgeons to interpret and could give misleading information.

Fig. 5.4
figure 00054

Image set from a second animal in the study acquired using both LSCI (a, b) and ICG fluorescence imaging (c, d). Time points are shown for baseline physiological flow in (a) and (c) and after a localized vessel occlusion is induced using photothrombosis in (b) and (d). LSCI images appear darker in areas of higher flow, while ICG images appear brighter in areas where the ICG dye is present. Scale bar corresponds to 0.5 mm

Full size image

One major advantage of LSCI is that it can be used to continuously monitor blood flow changes, both during baseline flow and during a clot formation. Figure 5.5 shows an example of a relative blood flow time course measured from the first animal in the study (same animal as Fig. 5.3). By calculating the correlation time within each region of interest and comparing to a baseline value, the relative blood flow changes can be calculated and easily visualized for four different vessel branches in the field of view. After the rose-bengal is injected, one vessel branch immediately clots, as shown by the sharp decrease in flow in region 4. The other vessel branches take a few minutes to fully clot, and then two of the branches re-perfuse briefly (regions 1 and 2), indicating that the clot broke free due to a pressure build-up. Because of the excess rose-bengal in the intravenous system, the clot reforms quickly and flow is again reduced in those regions. On the other hand, the ICG fluorescence provides the most useful information during the initial wash-in of the dye, which can be used to determine flow direction and separate arteries from veins. A reduction in maximum fluorescence reached during wash-in may also be used as an indicator of reduced blood flow, but quantitative results are difficult to interpret, especially if multiple injections of ICG are used during a single procedure.

Fig. 5.5
figure 00055

Temporal blood flow changes measured from the first animal in the study at four regions of interest in different vessel branches. (a) Location of each region in the camera field of view shown on a representative baseline LSCI image. (b) Relative blood flow changes for all four regions measured by changes in correlation time relative to a baseline value. Temporal changes over 40 min are visualized, during baseline flow, occlusion, and after the clot has formed. Arrows indicate the time points of the baseline ICG injection, the rose-bengal injection to trigger clot formation, and the post-occlusion ICG injection. Scale bar corresponds to 0.5 mm

Full size image

This initial study in a rat model has demonstrated that LSCI and ICG fluorescence angiography provide complementary information about vessel perfusion. ICGA has distinct advantages of being able to identify flow direction and differentiate between feeding and draining vessels. LSCI has the advantages that it does not require an exogenous contrast agent and can provide continuous assessment of blood flow during surgical procedures. LSCI also provides a more direct measure of flow, while ICG fluorescence truly indicates blood volume. This study has also demonstrated that both techniques can be performed simultaneously using a single light source, meaning that the LSCI images can be acquired simultaneously with ICGA simply by adding a second camera to capture the excitation light.

5.5 Intraoperative Imaging During Neurosurgery

To examine the utility of LSCI as an intraoperative imaging tool during neurosurgery, we performed a pilot clinical study (n = 10) using LSCI to image CBF during brain tumor resection surgeries. The goal of the study was mainly to assess feasibility of the technique for intraoperative use and to examine performance both under baseline flow conditions and during an induced change in blood flow. Because the hardware has been integrated into the surgical workflow, the technique is minimally invasive and the hardware required is simply an add-on to an existing neurosurgical microscope.

5.5.1 Clinical Procedure

In this pilot clinical study, ten patients were imaged using LSCI during brain tumor resection procedures at the NeuroTexas Institute in St. David’s Hospital, located in Austin, TX. The clinical study was approved by the Institutional Review Boards of the University of Texas at Austin and St. David’s Hospital. The microscope-integrated LSCI instrument shown in Fig. 5.2b was used for all patients. Before each procedure, the camera and laser attachments were added to the microscope and a surgical staff member covered the modified microscope with a sterile standard microscope drape. The hardware attachments did not interfere with the draping procedure or with the normal function of the microscope, meaning that the surgeon could use the microscope during the procedure as needed. The LSCI imaging procedure was performed at the discretion of the surgeon, either before or after the tumor resection portion of the procedure.

For all cases, the surgeon performed the craniotomy and exposed the cortical tissue before imaging. If the microscope was not already being used in the procedure, the surgeon positioned the microscope over the cortical area of interest for imaging, using the built-in illumination to guide positioning and placement. Because the LSCI hardware was integrated into the neurosurgical microscope, the surgeon could easily control the location and angle of the microscope head. This provided the flexibility needed to accommodate for the variability in the craniotomy location of each patient and enabled fast setup, with positioning complete in less than 5 min. After positioning the microscope, the built-in illumination was turned off and the laser diode was turned on to perform LSCI imaging. The surgeon used the focus and zoom controls built into the microscope head to prepare for imaging and flushed the cortical tissue area of interest with sterile saline to reduce specular reflections in the camera field of view. LSCI images were acquired for ~10–15 min, while recording the camera exposure signal and the patient’s ECG waveform simultaneously. An illustration of the microscope in use during surgery and the corresponding LSCI image is shown in Fig. 5.6.

Fig. 5.6
figure 00056

Adapted Zeiss OPMI Pentero neurosurgical microscope in use during neurosurgery. (a) The surgeon uses the microscope to assist with the tumor resection procedure. (b) The surgeon’s field of view as seen by our camera under the microscope’s built-in xenon lamp illumination. (c) When the surgeon is ready to perform LSCI, the xenon lamp is turned off and the laser diode is turned on. LSCI images can be visualized instantaneously, with a representative speckle contrast image shown (averaged over ten frames)

Full size image

Images of baseline blood flow conditions were recorded for all patients, in some cases over multiple tissue areas of interest. In two patients, the surgeon used bipolar cautery to perform surgical hemostasis within the camera field of view during LSCI imaging to induce a change in blood flow. After the surgical procedure, the LSCI hardware add-ons were removed from the microscope, leaving it exactly as it was before the surgery. Because the LSCI hardware is truly an add-on to an existing instrument, the system can be easily portable to different operating rooms and even different hospitals, as long as the surgical microscope is available.

5.5.2 Clinical Study Overview

The results from the pilot clinical study are promising and have demonstrated the instrument’s ability to visualize blood flow in a real time, minimally intrusive manner. For a subset of the patients, images were acquired with the microscope’s built-in color camera for comparison with the speckle contrast images. Figure 5.7 shows the camera field of view seen by the color camera under the built-in illumination along with corresponding representative speckle contrast images recorded from the same areas for three patients. The color photographs have been registered to match the orientation of the speckle contrast images, showing excellent alignment of the anatomical vasculature visible in the color images and the blood flow maps seen in the LSCI images. The blood vessels appear dark in the speckle contrast images, indicating that the vasculature in the field of view is unobstructed. The large vein running down the center of Fig. 5.7f may be the exception. The vessel does not appear to be obstructed in the corresponding color image, but the LSCI image indicates that this vein is flowing slower than the darker vessels nearby. The color images in Fig. 5.7c, e show blood pools present in the surgical field that obscure the view of the local vasculature. The corresponding LSCI images in Fig. 5.7d, f illustrate that the blood pools present in the surgical field do not interfere with image visibility, as flow is still clearly observed in the LSCI images where the blood pools are present. The ability to identify potentially obstructed vessels and to “see through” blood pools in the surgical field are examples of improved visualization possible with LSCI.

Fig. 5.7
figure 00057

(a, c, e) Color digital photographs taken from three different patients using the microscope’s built-in illumination and color camera. These photographs have been registered to match the orientation of the corresponding representative speckle contrast images (b, d, f). The speckle contrast images are averaged over ten frames. The bright white areas in both the color images and the speckle contrast images are regions of specular reflection. The camera field of view is ~2 cm × 1.5 cm

Full size image

5.6 Dealing with Motion Artifacts

One of the challenges involved with using an imaging technique that is highly sensitive to motion is dealing with physiological motion that does not originate from blood flow alone. The cardiac cycle leads to unavoidable pulsatile variation in CBF for any in vivo measurement. Using the recorded ECG signal for each patient, we can generate an ad hoc cardiac filter to reduce the blood flow fluctuations within each heartbeat. This allows for improved visualization of actual functional flow changes that may be occurring. Because the brain tissue is able to deform after a portion of the skull is removed during the craniotomy, there are also motion artifacts due to pulsation and respiration. This tissue deformation results in a constantly changing camera field of view, which makes it difficult to track a specific tissue region over time. Image registration was used to match the spatial location between images taken at different time points for a complete image set. This helped account for the physical motion of the brain visible in the original image sets and allowed more accurate tracking of tissue regions of interest over time.

5.6.1 Cardiac Filtering

To illustrate the magnitude of blood flow changes that occur within each heartbeat, the measured correlation time for a region of interest is plotted on the same time-scale as the ECG waveform recorded for the patient. A representative result from the first patient is shown in Fig. 5.8a, using the average correlation time measured from region 1 over a large blood vessel in Fig. 5.8c. This time course illustrates that the measured changes in \( {\tau_{\rm{c}}} \) are synchronized in time with the cardiac cycle. These pulsatile changes in blood flow can be tolerated during basic visual inspection of CBF and would not hinder use during applications where large changes in blood flow are expected, such as aneurysm clipping or vessel bypass. However, these fluctuations in CBF within each heartbeat may mask small changes in blood flow. This would be an issue during procedures involving functional mapping of the cortex, where activation may induce a change in blood flow as small as 10% from baseline values [29]. The magnitude of the fluctuations observed in Fig. 5.8a would limit the ability to identify small changes in blood flow, which is the main motivation for eliminating the artificial rise and fall of blood flow observed within the cardiac cycle.

Fig. 5.8
figure 00058

(a) The average correlation time within a region of interest co-localized in time with the ECG waveform recorded for the patient. This illustrates clear fluctuations in correlation time that are synchronized with the cardiac cycle. (b) The ad hoc ECG filter produced using the average correlation times from the same region of interest during the time required for 25 heartbeats. (c) A representative speckle contrast image showing the locations of two regions of interest. Region 1, located in the vessel, is used for (a) and (b), while region 2, located in the parenchyma, will be used for analysis in Fig. 5.9

Full size image

Because the fluctuations observed are synchronized with the heartbeat, we have designed an ad hoc ECG filter to reduce the beat-to-beat variability. This filter is similar to an adaptive filter technique for MRI data developed by Deckers et al. [45] and was first described by our group in Parthasarathy et al. [32]. The filter is designed to identify the time of the camera exposure relative to the ECG cycle and to account for the signal shape at that relative time. Because both the camera exposure signal and the ECG waveform are recorded during image acquisition, both the actual time of each speckle image and the actual time of each heartbeat are known. Thus, the correlation time measured from each speckle image can be assigned a “normalized time,” which is defined as the time of image acquisition relative to the nearest heartbeat. The normalized time is defined in (5.5)

$${t_{\rm{normalized}}} = \frac{{{t_{\rm{frame}}} - {t_{\rm{beat}}}}}{{{t_{{RR}}}}}, $$
(5.5)

where t frame is the time of the camera exposure, t beat is the time of the nearest previous heartbeat, and t RR is the duration of the heartbeat defined as the time between adjacent R peaks in the ECG waveform. Using the images acquired during the first 25 heartbeats in an image set, the average correlation time from a region of interest can be plotted against its normalized time to generate the ad hoc ECG filter function. An example for the same region is shown in Fig. 5.8b, where the blue circles are the measured \( {\tau_{\rm{c}}} \) values averaged from the region of interest in each image and the red line corresponds to the filter function generated by averaging the measured values with a large window moving average filter. Thus, each value of the filter function represents the average shape of the correlation time signal expected at a specific time in the ECG cycle. To apply the filter to the data, the normalized time of each image must be calculated. Then, the corresponding filter value for that normalized time is subtracted from the correlation time data point and the median correlation time value of the ECG filter is added back. To further reduce the noise in the filtered measurements, a small window moving average filter of width 0.05 s is applied to the ECG filtered result. The result after ECG filtering (red curve) and applying the small window moving average filter to the ECG filtered result (black curve) is shown for the same vessel region in Fig. 5.9a. This result illustrates that ECG filtering greatly reduces the variability within the cardiac cycle and gives a relatively smooth curve while preserving inherent physiological flow changes. One important detail to point out is that ECG filtering works very well for the specific region of interest used to generate the ad hoc ECG filter. However, the ECG filter shape varies greatly between different tissue regions and between patients, meaning that a general ECG filter cannot be used. Thus, each region of interest analyzed within each patient must have its own ECG filter shape to correct for beat-to-beat fluctuations in measured correlation time.

Fig. 5.9
figure 00059

(a) The ECG filtered result from the original dataset for region 1 over a large vessel in Fig. 5.8c. The red curve shows the ECG filtered output and the black curve shows the result after applying the small window moving average filter to the filtered output. (b) The ECG filtered result for the same vessel region after image registration has been applied to the image set. The registered filtered output is relatively flat, indicating that the registration process accounted for the large increase in \( {\tau_{\rm{c}}} \) observed in (a) between 14 and 16 s. (c) The ECG filtered result from the original dataset for region 2 in the tissue parenchyma from Fig. 5.8c, showing a slow oscillation in flow. (d) The ECG filtered result for the same parenchyma region after image registration has been applied to the image set, which is very similar to the original filtered output

Full size image

5.6.2 Image Registration

To account for large-scale tissue motion due to pulsation and respiration, image registration was performed in post-processing to align the camera field of view for each image set. Registration was performed using Elastix, an open-source software package for medical image registration based on the Insight Segmentation and Registration Toolkit (ITK) [46]. This program uses intensity-based image registration to adjust the position of a moving image to match a fixed image using the transformation type, similarity measure, and optimization procedure specified in a parameter file. The type of transformation used during registration is one of the most important parameters and must be chosen carefully. The transformation determines the types of deformations that are allowed to create the mapping between the fixed image and the moving image. Because the brain tissue is deforming, a nonrigid transform would be required for an exact mapping of the images to account for tissue distortion. However, LSCI records a 2D view of the 3D deformation and the microscope has a large enough depth of field to keep the image in focus during tissue motion. This limits the majority of tissue deformation in the recorded images to the x and y directions. Thus, a translation transform was used for initial analysis to preserve the image integrity and decrease the complexity of the image registration process. Mutual information was used as the similarity metric and adaptive stochastic gradient descent was used for the optimization procedure, as the developers of Elastix recommend both of these parameter choices for general medical image registration with good performance [46]. The speckle contrast images were provided to Elastix for registration, since the blood flow maps provided clear image features to improve registration performance. After image registration, the ECG filtering procedure was performed on the registered dataset and the filtered output was directly compared with the original results.

A side-by-side comparison of the filtered output from the two regions shown in Fig. 5.8c before and after registration is shown in Fig. 5.9. As shown by the fluctuations in the blue data points from the registered set in Fig. 5.9b, ECG filtering is still necessary as beat-to-beat variations are still clearly visible despite the image registration. Thus, the filtered results shown in the red and black curves were produced using an ECG filter function calculated using the first 25 heartbeats of the registered image set. Interestingly, the results from the region within the vessel shown in Fig. 5.9b indicate a relatively constant time course, while the result from the original dataset shown in (a) showed a large decrease in flow (increase in \( {\tau_{\rm{c}}} \)) between 14 and 16 s. After viewing the registered result, we can conclude that the decrease in flow was a result of the region of interest moving outside the vessel and into the surrounding tissue parenchyma, which has a slower flow than the vessel itself. This was confirmed in a video of the original image set with the vessel region of interest overlaid. The registered results from the tissue parenchyma region of interest shown in Fig. 5.9d appears remarkably similar to the original result in (c), indicating that the registration process did not have a large effect on the measured correlation time for the parenchyma tissue region. This suggests that the variation in blood flow within the tissue parenchyma where the capillary beds are located is higher than the variation produced from the tissue motion. Thus, there is a clear benefit to registering the images for regions of interest located in a vessel to ensure that the region remains inside the vessel throughout the time course of the analysis. This will eliminate any artificial changes in the time course due to movement of the spatial region into another tissue area, as observed with Fig. 5.9a. The benefit of image registration for analyzing tissue parenchyma regions is still being explored, but may be less important because of the inherent increase in variation observed in the parenchyma.

5.7 Conclusions

LSCI is a simple yet powerful tool for visualizing blood flow in real time with excellent spatiotemporal resolution and is ideal for use in a surgical setting. There is a clear need for monitoring CBF during neurosurgery and LSCI is a perfect candidate to overcome many of the disadvantages of currently available techniques. A direct comparison between LSCI and ICGA in an animal model demonstrated that the two techniques are complementary and provide different information to the surgeon. LSCI has the advantages that it does not require the use of an exogenous contrast agent, can be used continuously or repeated as often as needed, and provides a true measure of blood flow. The results from the 10-patient clinical study are promising and indicate the feasibility of using LSCI to monitor blood flow changes in a neurosurgical setting. The LSCI images showed excellent correlation with the anatomical vasculature captured by the microscope’s built-in camera and were unobstructed by the presence of blood pools within the field of view. The microscope-integrated instrumentation proved to be convenient and easy to use for the surgeons, providing real time, full field blood flow maps with minimal interference to the surgical procedure. Both the ECG filtering and image registration helped account for variability observed in blood flow as a result of the cardiac cycle and tissue motion. Future work will include a larger clinical study to assess LSCI performance during a wider variety of neurosurgical procedures, including cerebrovascular repair and functional mapping.