1.

Introduction

Systemic hyperthermia in the body is manifested as three separate entities: heat cramps, heat exhaustion and heatstroke, the latter two being life-threatening.^1–3 Signs of systemic hyperthermia include muscle cramps; excessive fatigue or weakness, or both; loss of coordination; a slowing of reaction time; headache; decrease comprehension; nausea and vomiting; and dizziness. Unlike heat cramps and heat exhaustion, heatstroke is a true medical emergency and is often fatal if not properly and promptly treated.^4–6 Heat stroke occurs particularly in non-acclimated individuals who are exposed to high environmental temperatures. The body normally generates heat as a result of metabolism, and the body is usually able to dissipate the heat by either radiation of heat through the skin or by evaporation of sweat. However, in extreme heat the thermal regulatory mechanisms fail, and body temperature rises. Usually, above 40°C (104°F) brain damage occurs and death follows if vigorous measures are not instituted.^7,8 Thus, early diagnosis and proper treatment are crucial for the patient’s survival. Immediate cooling and support of organ-system function are the two main therapeutic objectives in patients with heatstroke. Several cooling methods have been presented in the literature including immersion in water at different temperatures, evaporative cooling, ice pack application, pharmacological treatment and invasive techniques.^9–11 The populations most susceptible to heatstrokes are infants, older people in poor health (often with associated heart diseases, lung diseases, kidney diseases, or on certain medications that make them vulnerable to heat strokes), athletes, or outdoor workers physically exerting themselves under the sun. Susceptibility is greater in those with a disorder of the skin or sweat glands, and those taking anticholinergic drugs (which reduce sweating). Overstrenuous activity, unsuitable clothing, overeating, and drinking too much alcohol are sometimes contributory factors. However, some individuals can develop symptoms of heatstroke suddenly and rapidly without warning.

It is widely accepted that heatstroke develops as a result of an increase in body temperature leading to multiorgan damage. Therefore, methods to monitor and measure changes in internal temperature are important. Current clinical methods for monitoring temperature are divided into two categories, invasive and noninvasive. Techniques such as tympanic membrane thermometry, zero heat flow, microwave radiometry, magnetic resonance thermometry, and ultrasound thermometry besides the conventional oral or rectal electronic thermometers are in use.^12,13 However, accuracy, complexity, cost, portability, safety, and continuous measurements can influence the suitability of these techniques. For instance, rectal temperature may not accurately reflect brain temperature, especially during global ischemia. The development of a non-invasive device which overcomes these factors will be vital in medical management of heatstroke. Interest in using optical techniques is on the rise as they possess many advantages such as being noninvasive, portable, relative low cost, easy to maintain, can be used repeatedly over a prolonged period of time with no adverse effects, and offer both high spatial and temporal resolution.^14–16 So far, several researchers have successfully demonstrated the feasibility of using optical techniques such as near-infrared spectroscopy^12,17–19 optical coherence tomography,²⁰ laser speckle imaging,²¹ and fiber-optics sensors²² to monitor temperature changes in a noninvasive fashion from turbid media as well as from living tissue.

Diffuse reflectance spectroscopy (DRS) is an established tool widely studied and used to investigate physiology below tissue surface. It is a simple, low-cost, and noninvasive modality with great potential for diagnostic applications as it is sensitive to absorption and scattering properties of biological molecules in tissue. Therefore, it has the potential to provide biochemical composition (chromophores content) and morphology (structure) information about biological tissue.^23–28 The spectroscopy of tissue is possible owing to the weak absorption of the tissue in the near infrared (NIR) window (650 to 1000 nm). Classical DRS platforms include a light source, detector (spectrograph), and a pair of optical fibers separated by a fixed distance for delivery (illuminating fiber) and collection (collecting fiber) of light during measurements. Light entering the tissue undergoes a combination of multiple scattering and absorption and the reflected light from the tissue, called diffuse reflectance ($R_d$), is recorded by the detector fiber. The light intensity decreases during its ‘banana path’ propagation according to the concentration of the absorbing chromophores (hemoglobin, fat, and water).^29–31 It has been shown that separation between the source and the detection points determines the sensitivity of the measured reflectance to the absorption coefficient (large separation distances, $>\sim 10\mathrm{mm}$) and to the scattering coefficient (short separation, $<\sim 1\mathrm{mm}$).^32–39 Optical properties (absorption and scattering) can be extracted from the captured radiation with analytical models or Monte Carlo simulations. Even though there are several ways to model the measured diffuse reflectance spectra, in this work we used the two-source diffusion-based model for tissue reflectance as described by Farrel et al.⁴⁰ The use of diffusion theory is applicable here since: 1. measurements are obtained far from the boundaries, 2. the source-detector separation is $\sim 12\mathrm{mm}$, and 3. the absorption effects in brain tissue are less than the scattering (the brain medium is very diffuse).

As mentioned, several research groups have used NIR spectroscopy techniques to study the influence of temperature on NIR spectra since both photon path length and refractive indices are temperature dependent and because of the strong temperature-sensitivity of water absorption, which is reflected in intensity changes and wavelength shifts around its peak at 970 nm.^18,41–44 In addition, the effects of temperature on major tissue absorbers such as hemoglobin and water have been studied elsewhere.^45,46 Works were obtained also to measure glucose under temperature perturbation.^47–50 The aim of this study is to continue exploring this possibility in the wavelength range from 650 to 1000 nm during heatstroke with a different approach and configuration. We point out three differences as compared to the other techniques. First, we focus here on the temperature dependence of the absorbance spectra shift and its second derivative characteristics at specific wavelengths to evaluate the internal temperature of brain tissue. Second, in our setup the source and detector are orthogonal to each other on a curved brain surface with $\sim 12\mathrm{mm}$ separation. Finally, we assume that during changes in temperature there are changes in tissue concentrations such as hemoglobin and water. These molecules absorb specific wavelengths of light which alters the optical properties of tissue and consequentially alter the shape of the diffuse optical spectra. Therefore, we hypothesize that DRS can be used to monitor change in internal brain tissue temperature during heatstroke injury.

2.

Materials and Methods

2.2.

Instrumentation

A schematic diagram of the setup is drawn schematically in Fig. 1. The setup consists of a wideband light source, two optical fibers, spectrometer, temperature controller, pulse oximeter, rectal probe, and a computer station with control software. The lamp inside the light source (Stockler & Yale, Inc., M1000) was replaced with a broadband quartz-tungsten-halogen (EIKO, EXR) to gives a smooth broadband spectral profile along the NIR region and stable intensity over time. One of the optical fibers (illuminating fiber) is connected to this light source with a diameter of $D=5\mathrm{mm}$. The light emanating from the brain is delivered to the spectrometer (Stellarnet, EPP2000) via the second optical fiber (collecting fiber, $D=2\mathrm{mm}$) positioned perpendicular to the first fiber; That is, the diffuse reflectance was collected by the second fiber-optic probe that was mounted on the surface of the mice scalp at 90° to the brain surface and connected to a spectrometer (see Fig. 1).

Fig. 1

Schematic configuration of the experimental apparatus. Light exiting the skull is collected by the collecting fiber mounted at an angle of 90 deg with respect to the incident beam. Source-detector separation (arc length) is $\sim 12\mathrm{mm}$.

In this configuration, the reflected light collected by the collection fiber propagates through the brain layers (scalp, skull, CSF, gray, and white matter). But at our source detector separation the dominant signal is from photons passing through the brain. Hence the detected signal is sensitive to changes in optical property from the brain. The reflected signal is then dispersed and detected by the charge coupled device (CCD) array in the spectrometer. The spectrometer system has a 2048-elements linear CCD array with 12-bit resolution, and with a spectral range from 240 to 1200 nm. The spectrometer was controlled by a laptop computer running Windows operating system and is connected to the USB port of the computer. Spectrometer interface software was used between the spectrometer and the computer to display the optical reflectance in real-time. Both fibers were connected to a mechanical stage translation for position control and care was taken to ensure full contact between the probes and the brain surface throughout the experiments.

3.

Results and Discussion

Figure 2(a) shows the absorbance spectrum measured on a representative mouse brain (mouse #1, Table 1) with three different temperatures (32°C, 38°C, and 40°C) out of 10 measurements, conducted on the same animal, for easier visual representation. In Fig. 2(b) the region between 940 and 1000 nm was magnified to emphasize the change in absorbance intensity peak around 970 nm. In both panels, the effect of temperature on the absorbance spectrum is clearly observed and the difference between spectral intensities is noticeable. Based on the effect of temperature on the absorbance measurements shown we assume that both scattering and absorption alterations occur as a result of chromophores and morphological changes. The dominant feature is in the absorbance of water that peaks around 970 nm. As it can be seen, this peak decreased and shifts slightly to a shorter wavelengths (left shift) as temperature rises. This shift to a lower wavelength as the temperature increases resemble the Wien’s displacement law.⁵⁹ The intensity and shift changes at 970 nm are attributed to the fact that an increase in temperature results in a change of water absorptivity (and concentration) as a result of weakening and reduction of hydrogen bonds; water molecules absorb photon energy attributed to the vibrational overtone of the $\mathrm{O}─\mathrm{H}$ bond around 970 nm.^18,60 Therefore, the changes at this wavelength can be considered as a qualitative indicator of water accumulation in the tissue and therefore tissue and cellular edema during heatstroke injury. It is important to note the change occurring in the spectral shape of the absorbance to understand physiological changes during heatstroke. Specifically, we are interested in several wavelengths across the spectra that reflect tissue composition; i.e., characteristic changes in tissue composition at particular wavelengths could be indicative of chromophore behavior. The absorption coefficient of oxyhemoglobin differs from that of deoxyhemoglobin along most of the NIR spectrum. However, at a certain isosbestic wavelength ($\sim 800\mathrm{nm}$), the absorption curves cross each other and represent the same value. Thus change in reflectance at an isosbestic wavelength is affected mainly by blood volume (and hence, blood flow) while a change in a non-isosbestic wavelength is affected by the oxy-deoxy ratio as well. The absorbance around 650 nm is mostly affected by the change in deoxyhemoglobin while at 840 nm it is more affected by oxyhemoglobin. Therefore, one can consider the difference between the measurements at different wavelengths as a qualitative representation of the blood oxygenation level. Changes in absorbance intensity spectra confirm our hypothesis that heatstroke causes changes in the optical properties of brain tissue with resultant morphological damage and tissue composition changes. In order to highlight spectral differences that result from an increase in temperature, the spectral second derivative was calculated and is depicted in Fig. 2(c). The spectral region between 700 and 800 nm, magnified in Fig. 2(d), reflects the sensitivity to biochemical (specifically hemoglobin) variations around the center of 760 nm. The changes in absorbance around the isosbestic point (800 nm) suggest that cerebral blood flow is affected following induction of heatstroke. To evaluate the relationship between spectra reflectance and second derivatives versus temperature an exponential function were calculated and depicted in Fig. 3. As can be seen, as the temperature increases the peak reflectance at 970 nm decreases in an exponential fashion as a result of increase in absorption while the peak of the second derivative reflectance at 760 nm increases as a result of hemoglobin variations. We conjecture that heat diffusion to the tissue area may causes hemoglobin denaturation and thereby affect its ability to bind oxygen although the temperature is relatively low. The change in optical absorbance spectra of various hemoglobin species with temperature was also noted in earlier reports.^45,46 Figures 2 and 3 show a consistent pattern among the participant animals throughout this study and confirm our hypothesis that changes in NIR signals can reflect changes in internal temperature in injured brain. Simultaneous measurements of the body physiologic parameters were monitored during heatstroke concurrent to spectral measurements. The variations in heart rate and ${\mathrm{SpO}}_2$ versus temperature are indicated by the plot in Fig. 4. As shown, an increase in body temperature causes the heart rate to increase and concurrently induce a decrease in oxygen saturation until death. Please note that the ${\mathrm{SpO}}_2$ is generally above 90% because pulse oximeter measures the saturation of full-oxygenated arterial blood. However, ${\mathrm{SpO}}_2$ in Fig. 4 is quite low ($\sim 75\%$) even at normal temperature environment. We found that in three mice, the ${\mathrm{SpO}}_2$ was around 75% while the other two were around 85%. This low ${\mathrm{SpO}}_2$ maybe the result of oximeter deviation, or/and anesthesia effect. Since the mice are not ventilated, (over) anesthetization can cause respiratory depression and low ${\mathrm{SpO}}_2$. This is very difficult to precisely control in rodent experimental settings.

Fig. 2

(a) Representative reflected absorbance ($A_s$) spectra measured on the brain surface at three different temperatures out of ten using the setup shown in Fig. 1. (b) Enlarged region of Fig. 2(a) between $940\mathrm{nm}\le \lambda \le 1000\mathrm{nm}$ shows the decrease in the intensity peak around 970 nm as the temperature increases. A spectral shift of 10 nm from the baseline to a lower wavelength is demonstrated. (c) Absorbance second-derivative ($A_s^{\prime \prime }$) of 2(a). (d) Enlarged region of Fig. 2(c) between $700\mathrm{nm}\le \lambda \le 800\mathrm{nm}$ shows the change in intensity peak around 760 nm as temperature increases.

Fig. 3

Absorbance ($A_s$) and its second derivative ($A_s^{\prime \prime }$) intensity at 970 nm and 960 nm, respectively at different temperatures. These results are for the mouse of Fig. 2 and reflect the same trend for the other mice experiments. The points on the graph represent the discrete intensity measurement and the solid line represents the results of exponential fit to this data.

Fig. 4

Graph representation of heart rate (left axis) and ${\mathrm{SpO}}_2$ (right axis) versus temperature. Solid lines represent spline interpolation between data points. As the temperature increased from 40°C to 41°C the ${\mathrm{SpO}}_2$ is reversing its increasing trend, a point where the mouse died and heart rate start to decrease.

As seen from Fig. 3, and as we observed along this study, three factors are influenced by increase in temperature: 1. lower reflectance at $\sim 970\mathrm{nm}$ accompanied by 2. a spectral shift, $\mathrm{\Delta }\lambda$, from the baseline temperature and 3. an increase in its second derivative intensity at $\sim 760\mathrm{nm}$. These variations motivate us to define a following combined function coined reflectance-temperature index (RTI), which summarizes these significant elements of contrast in a single index. RTI defined as

RTI become more sensitive to temperature elevation as its numerator increase and the denominator strongly decrease causing a total increase in RTI. To bring RTI into a universal index (dimensionless) it was multiply by the average wavelength, $\langle \lambda \rangle$. Since the peak wavelengths which we determine the spectral shift, $\mathrm{\Delta }\lambda$, are not sharp but broad [Fig. 2(b)] we calculate the average wavelength, $\langle \lambda \rangle$, from the spectra band where the maximum is. Specifically, to determine $\langle \lambda \rangle$, we scan all the wavelength points where the maximum is and then we calculate the average. For example, in Fig. 2 for the 38°C, we found the maximum at the wavelet’s range of 978 to 982 nm. Averaging these five wavelengths results in $\langle \lambda \rangle =980\mathrm{nm}$. In continuation with Fig. 2, to determine the shift in wavelength as temperature increases ($\mathrm{\Delta }\lambda$) we found the maximum absorbance around 970 nm for each temperature and then we calculate the difference between. For example, in Fig. 2 the maxima of the baseline temperature (32°C) found to be at 980 nm while at 38°C it was found to be 970 nm which results in a 10-nm shift. From the entire study we found an average spectral shift from the baseline of $\mathrm{\Delta }\lambda =6.4\pm 1.93\mathrm{nm}$. The relationship between RTI and temperature derived from Figs. 2 to 3 through an exponential dependence of the form $\mathrm{RTI}=a\times \exp (b·\text{Temp})$ is display in Fig. 5(a) for the above specific mouse. The points on the graph represent the discrete RTI measurement and the solid line represents the results of exponential fit to this data. In Fig. 5(b) a plot of RTI versus temperature data for the entire experiments ($n=5$) is presented. The data in the graph are the mean and the error bars refer to standard error of positive and negative deviation separately, calculated over each temperature. The solid curve in this figure is the exponential fitting similar to what is shown in Fig. 5(a). As observed, increased temperature leads to a measurable increase in RTI, indicating the feasibility to follow temperature elevation in a non-contact manner using the DRS method. The exponential behavior we observed resembles the exponential increase of intracranial pressure (ICP) during brain injury although further study is necessary to determine the nature of the relationship. The central nervous system is very sensitive to hyperthermia that results in neurologic complications from injury to the cerebellum, basal ganglia, anterior horn cells, and peripheral nerves. The increase in absorbance at 760 nm (the dominant chromophore being deoxyhemoglobin) with rising body temperature can be explained by two phenomena. First, as a result of cardiopulmonary compromise⁶¹ systemic oxygenation of hemoglobin decreases (as evidenced by decreased oxygen saturation on pulsoximetry in Fig. 4), second, hyperthermic brain injury can result in increased ICP, consistent with previous observations.^62–67 It should be pointed out that no direct assessment of brain function (or dysfunction) was conducted. However, the changes in blood flow (800 nm absorption) provide indirect evidence of altered brain physiology. Neurobehavioral tests were not carried out during this experimental set up due to the terminal (end-point being death) nature of the experiments but can be studied in future nonterminal experiments.

Fig. 5

(a) Plot demonstrating the relationship of the reflectance-temperature index (rTI) to temperature for a single mouse of Fig. 2. As noted, an exponential relationship in the form of $a·\exp (b\times \text{Temp})$ was obtained. Here, $a=0.00084$, and $b=0.1$. (b) Graph showing RTI versus of Temperature from the entire experiments ($n=5$). $\overline{a}=0.00175\pm 0.0011$, and $\overline{b}=0.1143\pm 0.022$. Data are expressed as mean $\pm$ standard error.

As we show above, the relationship between RTI and temperature can be derived through an exponential dependence of the form $\mathrm{RTI}=a\times \exp (b·\text{Temp})$. From this expression the temperature can be extracted,

To bring this equation to a more general term and based on the entire experiments ($n=5$), averaged $a$ and $b$ were computed ($\overline{a}=0.00175\pm 0.0011$, and $\overline{b}=0.1143\pm 0.022$) and Eq. (6) was modified accordingly with $\overline{a}$, $\overline{b}$ instead $a$, $b$. Then, the modified Eq. (6) was used back on each of the five mice to predict the temperature and to compare it with the actual temperature. To obtain high correlation fitting, a $k$-factor was added to the modified equation. Hence,

To demonstrate the possibility of predicting the temperature ($T_{\mathrm{pred}}$) on a new mouse, not included in the previous experiments, we use Eq. (7) and temperature model prediction versus actual temperature is presented in Fig. 6 for the new mouse. A good correlation between prediction to actual temperature was obtained with $R=0.9$. Inspection of the correlation plots also reveals small RMSE and MRE, respectively. $k$ in this experiment was set to 5.75. The good agreement between the prediction and actual temperature data implies that: 1. RTI can be a candidate marker for temperature assessment during heatstroke and 2. temperature can be monitored noninvasively using the current DRS and Eq. (7). An important point to stress is that Eq. (7) is an empiric mathematical expression and should be studied in future works by increasing the number of animals and comparison with direct brain temperature measurement using conventional probes. This will be the next step in our heatstroke research.

Fig. 6

Comparison between temperatures obtained from Eq. (7) (model predicted) and from the YSI temperature probe measurements. The solid line is the best-fit linear regression fits to the data (open circles). The statistics were obtained from Eq. (5). High correlation between the predicted and actual temperature was obtained with $R=0.9$ and slope of 0.88.

As mentioned previously, optical absorption and scattering coefficients can be extracted with the use of the aforementioned Eq. (3). Since the diffuse reflectance $R_d$ is measured while $\rho$, $k_1$ and $k_2$ are known parameters we can find the absorption and scattering coefficients by iterating of the both coefficients until the calculated values of the reflection match the measured ones (Levenberg-Marquardt algorithm). It should be pointed out that the increased temperature may affect absorption and absorbance spectra differently since absorbance contains both absorbing effects caused by the chemical components and the scattering effects originated from the structural properties; while the absorption only reflects the chemical absorbing effects on light. The absorption and scattering behavior is illustrated in Fig. 7 at two ranges of temperature; baseline (32°C) and heatstroke (41°C) for the above representative mouse. The points on the graph of Fig. 7(a) represent the discrete calculated absorption coefficient at specific wavelengths and the solid line represents the spline interpolation fit to this data. On the other hand, in Fig. 7(b), the points represents the discrete calculated reduced scattering coefficient at specific wavelengths and the solid line represents the power-law fit to this data. Several studies have shown that the wavelength dependence of scattering in tissue in the NIR obeys a power-law dependence of the form,^68,69

Fig. 7

(a) Absorption, and (b) reduced scattering coefficients obtained by the use of Eq. (3) as a function of temperature for the representative mouse of Fig. 3. The solid line in (a) represent spline interpolation between data points and the solid line in (b) represent the results of power-law fit to this data based on Eq. (7). (c) Improvement in fitting of 7(b) after the last two wavelengths (950 and 970 nm) were taken out.

4.

Conclusion

Heatstroke is an abnormally elevated body temperature with accompanying physical symptoms including changes in the nervous system function. In this work, for the first time to our knowledge, we carried out experimental optical DRS on intact mice brains to a. assess the changes in tissue optical and hemodynamics properties that occurred during heatstroke, and b. to predict internal temperature noninvasively. The measurements were obtained with a single source-detector system orthogonal to each other, and the model of light propagation was employed to recover the optical property coefficients of cerebral tissue from the experimental data. During heatstroke, we note that the intensity peaks of the absorbance at 970 nm and in its second derivative at 760 nm correlate well with increased temperature through an exponential dependence. Reflectance-temperature index (RTI) that reflects changes in reflectance intensity and spectral shift from baseline was derived from DRS data. That is, we used the difference in spectra properties through RTI to estimate the influence of temperature on the internal rodent brain tissue. Although our pilot measurements are preliminary, results from a mouse show that RTI can be a good predictor for temperature measurement ($R=0.9$). It should be pointed out that the ability to determine the calibration factor $k$ in Eq. (6) is to have a complete dataset for calibration containing different temperature levels expected to be encountered during routine analysis. However, this can be accomplished during the calibration process with a dataset that is well representative changes in temperature conditions. Our future work will focus on further testing of the system on more mice and the applicability of Eq. (6) and applying appropriate calibration/validation modeling such as partial-least square (PLS) regression.¹² We also report the change in cerebral optical properties, based on the use of diffusion model, which reflects the variation in chromophores content and brain structure during heatstroke. The reported technique is simple, inexpensive, easy to implement, and can therefore be adapted in the future for temperature monitoring in clinical and laboratory settings as well.