1.1.

Spinal Cord Injury

Although there are different kinds of spinal cord injuries, in a generic sense, all spinal cord injuries begin with an unintentional physical event that disrupts cell membranes and tissue structures and causes materials that are foreign in uninjured tissue to contact and mix with healthy tissue and fluids.² Immediate cell death of neurons, glial cells, and endothelial cells occurs due to the mechanical trauma locally, defining the site of injury. This research investigates the chemical and physical state of injured spinal cord beginning within the first half hour of injury and extending to the subsequent 5 h. During this primary or immediate phase of SCI, a cascade of chemical and biological processes within the region of the injury is initiated and, regardless of the injury, the situation increases in complexity.

The secondary phase involves movement/migration of materials/cells in and out of the injured region and can last for hours to days. Note that the spatial distribution of the materials/cells during this phase may reflect the spatial distribution of the initial unintentional physical event and change over time. Observing the chemistry of injured cord tissue at the very beginning of the cascade might provide the best opportunity to contrast injured tissue with healthy tissue because healthy CSF is relatively low in protein relative to plasma. Protein produces significant background elastic and inelastically scattered light that may obscure scattering from SCI associated materials/cells. Thus, one main motivation for this research was to determine what spectroscopic information may be accessible during the immediate or primary phase. Any chance of success diagnosing and possibly treating SCI without physical contact depends on our capacity to detect and characterize chemical processes in turbid and delicate materials.

In addition to blood flow, which also distributes materials to/from the injured region from/to other parts of the body, our choice of timescale is comparable to that of the passive transport of molecules in the fluid media, i.e., CSF that fill the interstitial spaces within a healthy spinal cord. We hope that observing and possibly enumerating and identifying primary processes might suggest tactics and strategies to arrest a sequence that if left unattended will eventually form a glial scar. Alleviating scarring in turn could permit attempts at rehabilitating an injured spinal cord and promote healthier outcomes. Being able to predict that a glial scar will or will not form given the condition of a contused cord would itself be very useful in helping to decide, as soon as possible after SCI, whether desperate measures can or should be taken that entail no greater risk to the patient than doing nothing.

1.2.

Imaging and SCI

If the initial effects of the injury reduce blood flow into and out of the injured region, the resulting hypoxia will cause additional damage to the affected tissues.³ PV[O]H⁴ is simultaneously sensitive to (1) the presence/absence of blood and (2) the oxygenation state of the hemoglobin and so would seem to be advantageous. Other techniques will almost certainly be applicable to addressing related issues such as blood flow, e.g., Doppler-based optical techniques or optical coherence techniques.⁵ Flow-based measurements have been made in other parts of the body, e.g., femoral artery in the context of SCI but not in the cord itself. We address some of the practicalities of trying to apply optical techniques directly to the cord in the immediate aftermath of injury. Note that the possible utility of low level laser therapy (LLLT), also known as photobiomodulation (PBM) at the earliest times might also be coupled with and benefit from this exploratory study.⁶

Previously,² we described the chemistry and the subsequent chemical and physical events of SCI by studying ex vivo spinal cords in a rat model over a 4-day, 2-week, and 8-week post injury timescale. Near-infrared (NIR) probing revealed enhanced fluorescence that was associated with the injury. Thus, we might hypothesize that in vivo excitation of fluorescence during laser scanning might reveal the progression of the injury because PV[O]H has a well-defined response when fluorescence increases and decreases.

We suspect that diagnosis and treatment of SCI will probably involve many existing tools and techniques.⁷ Magnetic resonance imaging, computerized tomography, and x-rays are noncontact imaging modalities that give a three-dimensional, relatively high resolution view of the injured tissue without the need to surgically expose the cord, but none of these provide chemical information. Very high-resolution ultrasound (VHRUS) also produces clear images in vivo that are particularly rich in information regarding blood, but the cord must be exposed, in physical contact with a transducer, and physically/spatially stabilized in order to avoid motion defects. As for any technique in vivo, we have the inevitable blood flow and respiration-induced movement, and while it still provides valuable information concerning structural tissue damage and perfusion, or lack of perfusion, VHRUS does not contain chemical information. Even more invasive techniques in use today for research⁸ that may someday form the basis for an approach for treatment might benefit from real-time measures of physiological chemical signaling, i.e., real-time fluorescence changes. A major difficulty in attempting to obtain chemical information spectroscopically in vivo stems from the turbidity of even healthy spinal cord tissue.

In attempting to perform noninvasive in vivo spectroscopic probing for blood and tissues analysis⁹ in skin, we needed to deal with the turbidity of biological materials in general. To this end, we developed “PV[O]H” to quantify intravascular blood volume and composition changes. To obtain such information, one needs to deal simultaneously with both the spectroscopy and the propagation of light in the system, i.e., the turbidity. To date, we have validated PV[O]H in human and rat model studies involving skin, bacterial cultures in various media, and in unambiguous inanimate model systems.^10–13 The meaning/nature of the quantities calculated using PV[O]H depends critically on context. When applied to skin that has a capillary vascularization, PV[O]H calculates changes in the hematocrit (Hct) and total vascular volume of the capillary network.

It is essential to emphasize that in the context of imaging spinal cords with PV[O]H we are not suggesting that we are measuring blood when we calculate “apparent Hct.” We shall label the spinal cord images produced using values calculated by PV[O]H with the word “turbidity”, which is understood to be caused by the movement/migration of materials/cells in and out of the CSF in/near the injured region. The cells can be, e.g., T-cells or other lymphocytes and the materials can be, e.g., proteins such as immunoglobulins and other large molecules that scatter light elastically appreciably or perhaps fluorescence. Other types of molecules/materials, i.e., cellular debris, would have the same effect so more information may be required to identify these cells/materials. To this end, we note that NIR probing can be done simultaneously to implement PV[O]H while obtaining Raman spectra.

The algorithm itself is very simple and rests on assumptions that are met by the problem at hand, i.e., probing spinal cords in vivo. But the interpretation of the results of applying this algorithm is not as if it were being applied to capillary blood in skin. PV[O]H is extremely sensitive to (1) changes in elastic scattering and/or (2) changes in fluorescence/Raman yield for any reason in any tissues. Chemical information can often be inferred from fluorescence measurements utilizing endogenous or exogenous fluorophores.^14,15 PV[O]H can be used to produce images based on perfusion, Hct, fluorescence, or Raman spectra thereby potentially providing a new noncontact imaging modality not currently met by any other modality.

1.3.

PV[O]H Algorithm

Probing nearly any biological material with NIR laser light produces a remitted spectrum roughly as shown in Fig. 1. All the light that is collected caused by probing with incident NIR light has been either elastically scattered (EE) or is remitted with a wavelength shift, and so we call it inelastically scattered (IE). IE contains both Raman scattered light, phosphorescence, and fluorescence, and we use both together as shown in Fig. 1. EE and IE are produced by two fundamentally different processes, and so equations that describe their observation are independent.

Fig. 1

Inset: raw typical single 20 ms frame of CCD output using 200 mW of 830 nm excitation on a human fingertip. Bottom: same frame showing sections of emission integrated to estimate inelastic emission (IE, $\approx 500$ to $1750{\mathrm{cm}}^{-1}$) and elastic emission (EE, $-30$ to $+10{\mathrm{cm}}^{-1}$).

The volume fractions for RBCs, plasma, and “static tissue,” which we take for everything else in the cord, sum to unity implying that there are no voids. This is summarized in Eqs. (1) and (2) using $\varphi$ for each of the volume fractions, i.e., RBCs, plasma, and static tissue in the probed volume:

We used the radiation transfer equation (RTE)⁷ to propagate the incident light from air into a three phase, three layer medium in the single scattering limit and with the phases, i.e., the RBCs, plasma, and static tissue distributed homogeneously in the probed volume. We implicitly assume that RBCs and plasma inside the cord volume have the same optical properties as those freely circulating elsewhere. Since in skin we were able to obtain agreement with experiment by summing contributions to IE and EE linearly, and the scattering coefficients for skin are within a factor of 2 for those of cord tissue, and still at least one order of magnitude less than that for RBCs, we expect that we can write two independent equations:

From the definition of hematocrit (Hct), we have:

The six parameters in Eqs. (3) and (4) can be determined using the RTE and published scattering and absorption coefficients. Presently, our interest is better served by noting that since we have two linearly independent equations linking two measured quantities, EE and IE, to the two volume fractions, ${\varphi }_r$ and ${\varphi }_p$, for each simultaneous measurement of EE and IE, it is possible to invert Eqs. (3) and (4) to obtain

There are six parameters ($a,b,c,d,e$, and $f$) for which we must obtain numerical values. We can do this using constraints based on empirical data or assumptions. ${\mathrm{EE}}_0$ and ${\mathrm{IE}}_0$ are values obtained at the start of probing and all subsequent variation of EE and IE are with respect to these values. Thus, the apparent values of “${\varphi }_r$” and “${\varphi }_p$” so obtained are with respect to some arbitrary choice of starting values. In the present case, the device was calibrated in such a manner that the starting reference point is $\mathrm{Hct}=28.65$ and all figures and graphs presented herein will reflect that choice. Since we are interested in changes in the cords over time, the choice of starting point in this methodology exploratory study is arbitrary at this point.

The experimental apparatus we used for this work was previously calibrated empirically by monitoring capillary blood in human fingertip skin in vivo using dialysis-induced blood composition changes independently monitored by an FDA- approved gold standard device called the CritLine. The CritLine analyzes blood inside the dialysis machine to obtain a value for Hct and thereby ${\varphi }_r$ and ${\varphi }_p$.

3.1.

Line Scans

The variation of the raw EE and IE and the calculated associated “apparent” Hct using the PV[O]H algorithm as the laser is scanned back and forth across the same trajectory on a healthy spinal cord are shown in Figs. 6(a) and 6(a). In Fig. 6(a), we display the raw IE and EE, i.e., contained in successive 20 ms frames, after having applied a 25-point adjacent average smoothing procedure.

Fig. 6

(a) The EE and IE obtained while scanning a cord for the purpose of showing how this imaging works. This scan did not follow the protocol described in Sec. 2. In this case, the scan started at A with the rat motionless for 80 s, then scanned to B in 200 s, stayed at B for 80 s, scanned back to A in 200 s, stayed at A for 80 s, rescanned from A to B in 200 s, stayed at B for 80 s, rescanned back to A in 200 s, and collected last 80 s at A. The laser probing was at the location indicated in (b): (1) at beginning and end of experiment and (2) between the red lines in (a).

We observe the average IE to decrease monotonically with some fluctuations either increasing or decreasing. In contrast, the EE is relatively constant on average but does fluctuate in a manner different from, but clearly correlated with, the IE fluctuations. This is analogous to observing the IE and EE at a single location but as a function of time, with fluctuations caused by, e.g., variations in perfusion. Here, we observe fluctuations due to variations in surface topography, and the composition of the cord is translated under the probe laser.

The bleaching can be inferred by the rise in “apparent Hct” at the hold locations.¹⁰ By the end of the experiment, the bleaching at position A, i.e., the amount of Hct, increases while holding at A is much less than at the beginning of the experiment also while holding at A for the same time interval. This behavior is exactly what we observe in skin with the bleaching being complete at a single location in about 5 to 10 min of continuous probing at these power levels.¹⁰ Once fully bleached, the tissue remains bleached for at least 5 h in actual live tissue, without additional exposure to the laser.^14,15 This qualitative behavior of the probing/algorithm applied to spinal cord is essentially identical to that when applied to skin and despite the fact that we calibrated the algorithm in use against capillary blood Hct variation in skin, the same set of parameters $a$ to $f$ for Eq. (5) produces acceptable variation for image formation.¹⁰ That is, we do not expect the actual apparent Hct values to be meaningful, but the relative variation of “apparent” Hct as employed for imaging should be systematic, reproducible, and in fidelity with the actual physical appearance of the cord.

For comparing injured versus healthy spinal cords, we performed line scanning as described in Sec. 2 according to Fig. 3. These line scans of apparent Hct over time allow a methodical recording of fluorescence changes, increases in apparent turbidity, e.g., as might occur with bacterial infection of the CSF or the occurrence of excess protein in the CSF. The height of the surface of the animal changes less than the angle of incidence varies with the surface topography of the cord. Variation in distance between the laser aperture and the cord surface affects the spatial resolution of the image by changing the laser spot size. Visual observation suggests that in this study that variation is very small, probably less than 10%, i.e., the spot diameter might vary by between about 100 and $110\mu \mathrm{m}$ diameter. Inflammation, edema possibly, and other effects could induce similar changes over the course of time.

At each hold location, we observe the Hct to increase, regardless of location, or whether the cord was injured or not, until the tissue is completely bleached. While in motion between locations, the Hct initially always decreases and then a variety of Hct responses are observed, including very sharp and deep dips such as near 1200 s in either the preinjured or first postinjury scans in Fig. 7. The Hct tends to decrease in between the reference locations because the tissue along the scan route bleaches more slowly than the tissue that receives continuous probing, i.e., the hold locations.

Fig. 7

All 11 line scans from pre- to postinjury in a single SCI experiment. The time scale for scanning indicates the progression of the SCI, the bleaching process, and the position of the scan, because of the simultaneous bleaching that is occurring. Note that this is an “injury” experiment and “control” experiments were performed in exactly the same manner, including total time intervals, except no impaction/injury was performed. The preinjury scan is in the back with the successive scans in front. This is plotted so that the earliest scan is displayed translated slightly to the right and the later scans extend slightly to the left. Semi-log plots of IE for all scans at any particular Raman shift are essentially linear over the initial 5 h post-SCI including the preinjury scan.

The IE (total counts at Raman shift $1078{\mathrm{cm}}^{-1}$ including underlying fluorescence) collected for all 33 measurements were plotted semilogarithmically in Fig. 8. Despite the fact that the measurements combined results from three different hold locations, i.e., A, B, and C, the linearity of the result indicates that there is an exponential decay in IE during the data collection in the hold at each location, whereas the EE does not change appreciably. This reinforces our inference that autofluorescence photobleaching occurs and indicates that the spatial registration of the scanning system is reproducible and precise to within plus or minus the diameter of the laser spot $\approx 100\mu \mathrm{m}$.

Fig. 8

Semi-log plot of average IE for each of the 11 scans over the initial 5 h post-SCI including the preinjury scan.

The same data were collected over the progression of the SCI initiated event; all 11 scans for this particular experiment are shown in Fig. 8. Note that as the number of successive scans increases the increase in Hct at each hold position diminishes indicating that whatever material is being bleached, it is not replaced in the time interval between successive scans. Thus, while there is almost certainly some emission from blood, i.e., from both RBCs and plasma, the bleached material is not blood since as long as the heart is beating, the blood is totally and continuously replaced in seconds from the probed volume.

The precipitous dip near 1200 s is unusual in that an equally precipitous rise quickly follows. The strong localization of the apparent Hct variation forces us to consider the possibility that we scanned over the blood vessel that can be seen in Fig. 4(a). We observed this type of structure in the PV[O]H line-scan image repeatedly, suggesting that PV[O]H images might have good fidelity with the actual appearance of the cord in each case. While we observe that the conditions for straightforward application of PV[O]H are usually met, as a localized event, scanning over a superficial vessel with relatively large vascular volume and Hct might not meet the requirements. Compared to other tissues, this probing would provide a unique mutual variation in EE and IE that is inconsistent with the PV[O]H assumptions for applicability, i.e., Eqs. (1) and (2) are obeyed and elastic scatterer density beyond the single scattering/linear propagation range.⁹ This could result in an uncharacteristic calculated value for turbidity/apparent Hct at those locations. Related to that consideration, we should also be cognizant that direct laser irradiation of blood in the intravascular space could result in excessive localized heating and ultimately, tissue damage.

Figure 9 shows the “average turbidity” line scans calculated by PV[O]H as a function of position(time) averaged over three separate injured rats and three separate control, i.e., uninjured rats and averaged over all scans in an experiment, i.e., sequence number. The sharper quality of the “dips” observed for the control animals as compared with the broader dips observed for the injured animals may result from registration errors in the scanning procedure. All animals were initially scanned before injury regardless of whether the animal was ultimately part of the injured or control cohort. The process of inducing injury did not involve additional movement to use the impactor specifically to avoid the opportunity for spatial registration error. So despite the small number of animals in each cohort, it is also possible that the difference in the average images results from cardiovascular changes induced by the injury itself.

Fig. 9

Line scans for averaged over three injured and three control animals. “Turbidity” is apparent Hct calculated from EE and IE using PV[O]H calibration. The injured animal scans display more bleaching effects while the control animals displayed more narrow and well-defined “dips.” Note: 1 hemocycle corresponds to 3 s.

3.2.

Two-Dimensional Imaging

In order to explore the feasibility and possibly justify conducting a larger study, we performed a series of 2-D imaging experiments using one control and one injury model rat. The orientation and position of the surgical field and exposed spinal cord are shown in Fig. 10 along with an actual 2-D image. The orientation of all line scans, e.g., in Fig. 7 is shown in the black line and dots. The appearance in the images with rising and falling apparent Hct is independent of the position of scanning a specific location in the whole scan indicate that the choice of scan direction and total laser exposure per scan avoids photobleaching artifacts.

Fig. 10

Image of surgical field including spinal cord based on Hct values obtained in a scan trajectory [Fig. 4(b)] designed so that no point is probed twice so as to avoid photobleaching effects from an early test scan causing systematic artifacts in later images. Note the systematic fall off of the apparent Hct on the edges of the image due at least in part because of topographic variation, i.e., curvature. The spatial scale is indicated on both axes using the time from the beginning of the first scan in seconds since the actual distances are as shown in Figs. 4(a) and 4(b), and the possibility of photobleaching artifacts is determined by the time duration of the laser exposure.

Successive images are shown in Fig. 11 for an injured and a control rat. There are notable differences but a definitive comparison must await additional data to possibly show statistical significance. Presently, we observe that the images of the probed regions of both animals may reflect the effect of injury. A larger apparent Hct usually involves increased IE and decreased EE. A topographical origin of such an observation could occur if there is greater curvature in the injured cord than in the control cord. Note that if incident probing, i.e., laser light, was maintained at a location of high apparent Hct for too long, bleaching would be observed if such was the case. As another possible cause, if there was swelling in the injured cord this would be observed.

Fig. 11

Successive images of (a) an uninjured cord and (b) an injured cord and at various times after surgery to exposed cord. The apparent Hct scale is the same for each image and the orientation and size of each image is as indicated in Fig. 10. Note that injury is not induced until after the first 30 min scan.

Simulations based on the optical properties of cord tissue might reveal quantitatively how much change in apparent Hct could be associated with either effect. At this point, we cannot settle this question but we are encouraged that the images may in fact be classified based on the difference between internal physical and chemical status and topographical variation. It seems quite likely at this point that with more experience such difference could then be interpreted in terms of inflammation, edema, and/or swelling as one group of possibilities and bacterial infection or protein increase in the CSF as another. We note in passing that Raman spectra can be obtained at any location simultaneously with obtaining EE and IE and that would provide additional information to help classify and differentiate the possible causes and effects manifested in these images.

We note that the preinjury images of the two animals are essentially identical. And regardless of the above ambiguity, the successive images suggest that injury initiates processes that evolve over a period of $>2\mathrm{h}$ from the time of injury. The images for the control animal are essentially constant over the entire experiment arguing that hydration is adequate and that the spatial scanning is also executed consistently.

Analogous to the line scans Fig. 12 shows the average 2-D image from the sequence shown in Fig. 11 as well as the standard deviation of each pixel. The standard deviation of a pixel in the center of the images is smaller by about a factor of 3 than the difference of the apparent Hct between control and injury in the same location.

Fig. 12

(a) Average of spinal cord images from Fig. 11 of both (left) uninjured and (right) uninjured cord over scanning time. (b) Standard deviation of spinal cord images of both (right) injured and (left) uninjured cord over scanning time.

We find the 2-D images to be remarkably consistent with the line scans. The average preinjury line scans start at a higher turbidity than the first 2-D image regardless of whether the rats are control or injured group. However, this is to be expected since care was taken throughout the 2-D scanning process to limit the exposure of the tissue to laser-induced bleaching. On the other hand, the line scans started at point A with a full 5 min exposure to laser probing, which caused photobleaching and the expected increase in measured turbidity. Furthermore, we note that the average 2-D images in Fig. 12 clearly show slightly greater apparent Hct, i.e., turbidity for the injured animal over the control animal averaged over all locations and timeframe. We also note that later 2-D images show turbidity increase near the edges at the later times, possibly because there was some drying out of the tissues. Uneven drying might also make the standard deviations of the calculated turbidity larger at longer times.