1. Introduction

Medically induced injury to a nerve, termed iatrogenic nerve injury (INJ), has been reported in many procedures, and it results in adverse outcomes for both patient and practice [1]. INJs occur in 60-84% of challenging surgical cases and account for up to 25%, 60%, and 94% of sciatic, femoral, and accessory nerve lesions, respectively [1–6]. Overall, neuropathic pain is reported by 10-40% of patients after standard surgical procedures, with 2-10% experiencing severe chronic pain [7,8]. Since symptoms can cause severe pain with high morbidity [5,8–15], INJs often lead to significant medicolegal consequences, with compensation rewarded to patients in over 80% of certain cases [16]. To avoid INJs, standard care relies on a surgeon’s anatomical knowledge of the nervous system, visual acuity, and experience, as well as surgical exploration to identify nerves [1,17] after which the gold standard of intraoperative nerve monitoring can be utilized [15]. However, interpatient variability in nerve anatomical locations and distributions along with bloody surgical fields can prohibit effective intraoperative nerve localization, extending procedure times and increasing the risk of INJ [18–20]. To localize nerves before a procedure, magnetic resonance or ultrasound imaging can be employed [1,21–23], but these techniques cannot be used to view nerves during surgery and have severely limited nerve contrast [21]. For these reasons, there is a need for a tool that can enhance nerve contrast intraoperatively and in real-time to assist surgeons in rapidly identifying nerves for subsequent intraoperative nerve monitoring.

To address the inadequacy of current methods, multiple groups have turned to optical techniques for identifying nerves due to their potential for real-time intraoperative deployment. For example, spectroscopic methods, such as probe-based diffuse reflectance spectroscopy (DRS), have been applied for intraoperative nerve sensing [24,25]. Balthasar et al. used needle-based diffuse reflectance spectroscopy (DRS) between 500-1600 nm to identify nerves for nerve block applications [26,27]. Optical imaging-based methods have also been employed to enhance nerve contrast. For example, Gibbs et al. have developed fluorescent contrast agents to highlight nerves [28,29]. Other groups have used polarized light to enhance nerve contrast based on the intrinsic anisotropic structure of nerves [30,31].

Despite the growing number of optical techniques being applied for peripheral nerve localization and visualization, the optical properties of peripheral nerves and their relation to those of surrounding tissues are not well characterized. These properties, namely the absorption coefficient, reduced scattering coefficient, and scattering anisotropy, are needed to optimize optical-based nerve detection systems. Knowledge of these optical properties would assist in deciphering the mechanisms behind intrinsic DRS-based nerve contrast and could be used to optimize buried nerve detection and simulate various light illumination and collection geometries. Few studies, however, have reported peripheral nerve optical properties. One study determined the absorption coefficients of human sciatic nerve and surrounding fat from 300-2300 nm using an integrating sphere and applying the Kubelka-Munk equations [32]. Although helpful in deciphering nerve/fat contrast, this study did not report scattering properties and assumed isotropic scattering to calculate the absorption coefficients. Another study reported the absorption and reduced scattering coefficients in the visible (VIS) spectrum for human facial nerves and surrounding tissues [33]. Overall, previous studies have lacked stringent validation of optical property coefficients and have left near-infrared (NIR) and shortwave infrared (SWIR) optical properties only partially characterized.

To address the insufficiency in reported peripheral nerve optical properties, the primary goal of this study was to determine the absorption and reduced scattering coefficients of nerve and surrounding tissue in both rats and humans for a broad range of wavelengths, namely 352-2500 nm. The secondary aim was to use the optical properties to identify a spectral region that maximized photon penetration depth while maintaining nerve contrast. Once accomplished, a hyperspectral imaging system covering the optimal spectral region was employed to collect in vivo data for identifying specific wavelength combinations with the greatest potential for real-time embedded nerve imaging.

2. Methods

2.1 Absorption and reduced scattering measurement and validation

Optical properties were determined from reflection and transmission measurements along with the inverse adding-doubling (IAD) method. A dual-beam 150-mm diameter integrating sphere spectrophotometer (Cary 7000 Spectrophotometer with a diffuse reflectance accessory [DRA-2500, 99% Spectralon inner surface coating], Agilent, Santa Clara, CA, USA) was used to measure sample reflectance and transmittance from 352-2500 nm (Fig. 1(A)). A photomultiplier tube (PMT) collected light from 352-799 nm (2-3 nm bandwidth excitation) and a Peltier cooled lead sulfide detector collected light from 800-2500 nm (8-9 nm bandwidth excitation). The measurement protocol was as follows; first, a 99% Spectralon diffuse reflectance standard (Labsphere, North Sutton, NH) was placed in the sample and reference port to normalize the sample beam intensity to the reference beam intensity. Next, dark noise measurements were made by removing the sample in the reflectance sample port (13.5 mm diameter) for reflection dark noise (R₀) and by blocking the transmission sample port (13 mm diameter) for transmission dark noise (T₀). Total reflection (R_s | includes specular reflection) and total transmission (T_s) were measured with the samples in their respective ports. The measurements were normalized to the reference beam reflection off a 99% reflectance standard (r_std, t_std) with the sample in place, which provided the total reflectance (r) and transmittance (t) [34].

As an additional step, losses from glass absorption and specular reflection were corrected via measurement instead of using assumed glass properties. A glass slide was placed in the reflection port and a 99% reflectance standard was placed immediately behind the glass slide. A total reflectance measurement (r_t,g) was made to account for double-pass glass absorption. A light trap was then inserted to eliminate specular reflectance and measure only the diffuse reflectance (r_d,g), to account for both double-pass glass absorption and specular reflection losses. The correction factor for glass absorption was defined as c_a= 1/r_t,g and the correction factor for glass absorption and specular loss was defined as c_a,s= 1/r_d,g. Uncorrected total sample reflectance measurements were multiplied by the glass absorption correction factor and uncorrected sample total transmittance measurements were multiplied by the glass absorption and specular loss correction factor, which provided the final sample reflectance (r_s) and transmittance (t_s).

 

Fig. 1. Systems utilized for optical property determination and shortwave infrared (SWIR) hyperspectral imaging. (A) A dual-beam spectrophotometer collects absolute reflectance and transmittance via real-time normalization of sample reflection to a 99% diffuse reflectance standard. A photomultiplier tube (PMT) collects photons from 352-799 nm and a Peltier cooled lead sulfide (PbS) detector collects photons from 800-2500 nm. (B) Unscattered transmittance is measured by passing a beam through a sample and isolating unscattered forward propogating photons using a pinhole and detector at a sufficient distance. (C) A SWIR (1000-1700 nm) hyperspectral camera acquires images of a rat scaitic nerve and surrounding tissue, with an incandecent source utilized for illumination.

Download Full Size | PDF

As a final step a three-point moving average filter was applied to the r_s and t_s spectra to diminish noise.

The inverse adding doubling (IAD) method was then applied to determine absorption and reduced scattering coefficients from r_s and t_s [35]. IAD works by iteratively solving the forward adding-doubling algorithm, which uses a given albedo (α), optical thickness (τ), sample thickness (d), and anisotropy (g) to numerically solve the radiative light transport equation for r_s and t_s. These properties are iteratively adjusted until the measured r_s and t_s match those calculated by the adding-doubling method and are then used to determine the absorption (µ_a) and reduced scattering (µ_s$^{\prime}$) coefficients.

The refractive index and the anisotropy were set to literature values for each respective sample and the number of glass slides was set to zero in IAD since losses from the glass slides were accounted for via correction factors. A validated two-stage IAD approach was then employed to correct the miscalculation of optical properties in spectral regions of significantly high absorption compared to scattering [36,37]. For this method a power-law Mie-Rayleigh scattering equation

The optical property determination method was validated prior to tissue measurements. To verify accurate reduced scattering coefficients and scattering anisotropy factors, 1.1-µm polystyrene microspheres (Fisher Scientific 5100A) suspended in deionized (DI) water were measured. These microsphere solutions have known scattering properties given by Mie theory. Water, a well-characterized chromophore, additionally provided absorption coefficient validation. Solutions of 0.54% and 4.78% v/v polystyrene spheres in DI water were pipetted into a 0.96 and 0.46 mm pathlength cuvettes, respectively. The aforementioned measurement and processing protocol was used to determine the optical properties. The refractive index was set to 1.33 for water and the anisotropy was set to 0.9 in IAD [39]. The measured optical properties were then compared to Mie Theory or the literature for validation.

2.2 Anisotropy factor determination

Unscattered transmittance was measured to determine the scattering anisotropy factor. The 0.54% v/v polystyrene microsphere solution used for coefficient determination was employed for anisotropy factor validation. The solution was placed in a cuvette with a short pathlength (∼200 µm) to maximize unscattered transmittance over the scattered transmittance background. The VIS (352-799 nm | 3-4 nm bandwidth) or IR (800-2500 nm | 8-9 nm bandwidth) light impinged on the sample and was collected at a 130-mm distance by a 9-mm diameter aperture (0.0038 steradian solid angle) in front of a photodiode (Fig. 1(B); Universal Measurement Accessory, Agilent, Santa Clara, CA, USA). Sample measurements (T_u) were normalized to a measurement without the sample in place (T) after subtracting the dark noise measurement made with the beam blocked (T₀), which provided the unscattered transmittance (t_u)

To account for losses due to glass absorption and specular reflection the sample transmittance was multiplied by the aforementioned c_a,s correction factor (t_u,c= c_a,st_u). The corrected unscattered transmittance (t_u,c), the previously determined sample absorption coefficients, reduced scattering coefficients, and the pathlength (d) were substituted into Beer’s Law to calculate anisotropy (g) [40].

This measurement and processing protocol was subsequently employed for tissue characterization.

2.3 Ex vivo tissue optical property measurements

The optical properties were determined for human and rat nerves, fat, muscle, and tendon, which commonly surround nerves in vivo. Specifically, adult Sprague-Dawley rat sciatic nerve, abdominal muscle, tail tendon, and subcutaneous fat (IACUC # M1600085-01) were chosen for their optimal size. Cadaver antebrachial nerves, muscle, tendon, and subcutaneous fat (obtained through the Vanderbilt Cooperative Tissue Network, Nashville, TN) were extracted from readily available thawed cadaver arms. Rat tissues were extracted immediately after euthanasia and were stored in airtight vials with a Kimwipe soaked in phosphate buffered saline (PBS) at 4°C until reflectance and transmittance measurements were performed. On the other hand, human tissues were stored at -80°C after dissection. All samples were brought to room temperature before conducting measurements. Following extraction of the tissue from its vial, PBS was promptly applied to the tissues surface, and it was placed between two glass slides to prevent dehydration. Excess PBS was carefully drained. The drops of PBS also served to create consistent contact between the glass slides and the specimen. Spacers ranging from 0.23 to 0.73 mm were utilized to avoid tissue compression from the sample holder and ensure a constant tissue thickness. If the tissue specimen was too large for the spacers, a scalpel was utilized to cut out smaller pieces of the specimen. For measurement, the beam size at the sample was set to ∼2 mm² and was always <6% of the specimen area, which ensured the collection of radially scattered photons. The specimen area varied from ∼64 mm² to over 195 mm², covering 45-100% of the sample port. Diffuse light loss, when the port was not completely covered, was inherently corrected via the dual-beam approach [34]. Reflectance and transmittance spectra were collected in 4 nm steps between 352-2500 nm from 1-4 spots on the specimen, depending on its size and inhomogeneity. One nerve, muscle, fat, and tendon specimen were measured from each of the five rats and five human cadavers. After measurements were complete, the processing protocol, validated using the polystyrene sphere solution, was applied to the tissue measurements, with the assumed tissue index of refraction set to 1.4 and the anisotropy set to 0.9 [38]. In single measurement cases where IAD did not converge at a certain wavelength due to insufficient transmittance (e.g., ∼1936 nm water absorption peak), the resulting optical properties at that wavelength were considered unreliable and excluded from the mean calculations. The absorption and reduced scattering coefficients of the 40 specimens were averaged for each tissue type and model, and nerve optical properties were compared to muscle, fat and tendon respectively using two-tailed t-tests. Lastly, the effective penetration depth (δ_eff)

The scattering anisotropy was additionally determined to better inform Monte Carlo simulations used in identifying an optimal spectral region for nerve contrast. Unscattered transmittance was measured from a muscle, fat, and nerve tissue sample for four rats. Notably, unscattered transmittance could not be reliably measured from tendon due to its comparatively high scattering. These tissues were stored at -20°C to avoid degradation, as T_u was not measured within a few days after tissue extraction. A small segment of each tissue was cut to fit between glass slides with ∼200 µm spacers. Unscattered transmittance was collected in 4 nm steps from 352-2500 nm with the same experimental measurement protocol used in the polystyrene validation measurements. The validated processing protocol for determining anisotropy, established using the polystyrene microsphere solution, was then applied to the tissue measurements. The scattering anisotropy spectra of the 12 tissue samples were averaged for each tissue type.

2.4 Monte Carlo simulated diffuse reflectance

The diffuse reflectance of each tissue type was simulated from the determined optical properties using Monte Carlo eXtreme [41] to narrow down a spectral region of optimal diffuse reflectance-based nerve contrast. A generic light excitation and tissue geometry was set up wherein a 2-mm diameter flat top beam impinged on an 8 × 8-mm tissue cross-section with 50 × 50 × 50-µm voxel size. The modeled thickness of each tissue was set based on the thickness of the tissue specimen after extraction to approximate in vivo thickness. Muscle was defined with a 3-mm thickness, fat 2 mm, tendon 1 mm, and nerve 1 mm. At each wavelength, a simulation was run with 10⁷ photons using the rat and human absorption and reduced scattering coefficients, respectively. The measured rat tissue anisotropy factors were used for rat and human tissue simulations, with rat nerve anisotropy used for tendon, and the refractive index was set to 1.4 [38]. The total diffuse reflectance (calculated from total photon weights reemitted from the surface) for each tissue was normalized to the diffuse reflectance from a 2 mm thick slab with µ_a = 0 and µ_s$^{\prime}$ = 1000 mm^-1. This 2-mm slab approached maximum possible reflectance for the set geometry and was used to simulate a 100% diffuse reflectance standard. The modeling was performed using an NVIDIA GeForce RTX 3060 on an i7-11800H @2.3 GHz. The resulting simulated diffuse reflectance spectra and the photon penetration depth data narrowed down a promising spectral region for diffuse reflectance-based embedded nerve contrast. A hyperspectral imaging camera covering this optical range was then employed to verify the results.

2.5 In vivo shortwave infrared hyperspectral imaging

Hyperspectral reflectance images were acquired from an in vivo animal model to identify specific wavelength combinations within the previously identified spectral region that best enhanced nerve contrast. Adult Sprague-Dawley rats (n = 6 male, n = 5 female, 250-350 g) were anesthetized via isoflurane (3%, 3 L/min) inhalation and maintained under sedation (2-2.5%, 1.5 L/min) to perform a sciatic nerve preparation (IACUC # M1600085-01). A ∼3-cm incision was made from the gluteus muscles to the popliteal region exposing the biceps femoris, sciatic nerve, and surrounding fat. A SWIR hyperspectral imaging system (HinaLea Model 4440 SWIR 512 × 640p, HinaLea, Emeryville, CA, USA) with a variable focus lens (ZOOM 7000, Navitar, Rochester, NY, USA) was mounted ∼15 cm above the nerve (Fig. 1(C)). A 75W incandescent small animal heat lamp (Zoo Med Laboratories, San Luis Obispo, CA, USA) was used to illuminate the field of view at a ∼22-cm distance with an average exposure of ∼31 mW/cm² at the surgical field. Images were acquired with a 10-15-ms integration time in 5 nm steps from 1000-1700 nm. Each hyperspectral cube (I_s) underwent dark noise subtraction (I_b) and was normalized to the mean spectrum of a cube acquired from a 99% Spectralon diffuse reflectance standard (I_r | Labsphere, North Sutton, NH, USA) to yield the following reflectance cube:

Principal component analysis (PCA) was employed to identify wavelengths that provided optimal in vivo contrast of nerve using spectra from the hyperspectral data cubes. Before PCA, each pixel’s spectrum was normalized to its area under the curve (AUC) to disregard total intensity variability and focus on changes to the spectral shape. Principle component (PC) coefficients provided wavelengths contributing maximally and minimally (coefficient of zero) to the overall variance. Reflectance images at the wavelengths of maximal, minimal, or zero PC coefficient values were divided on a pixel-by-pixel basis to produce reflectance ratio images and assess the enhancement of nerve contrast.

To determine the depth at which embedded nerve could be detected, diffuse reflectance image cubes were acquired with sectioned 200 or 300 µm bacon fat and muscle slices iteratively stacked on top of the nerve. Images were then ratiometrically analyzed at the optimal identified wavelengths. The absolute difference between the ratiometric signal of the nerve region (R_{Nerve Region}) and the ratiometric signal of the fat or muscle region (R_{Non-Nerve Region}) in the image was calculated to quantify the nerve contrast (NC).

The NC was then fit to an exponential decay function with least-squares fitting and normalized to the uncovered nerve (0 µm) NC from the baseline offset. The nerve detection depth was defined as the depth where the NC diminished to 1/e. Regions of the image containing specular reflections, were not considered in the analysis. Specular regions were identified as having 1450-nm (water absorption band) intensities three standard deviations above the mean pixel intensity of the image.

3. Results

3.1 Optical property coefficient validation

The optical property determination protocol was validated using the well-characterized polystyrene microspheres in water solutions (Fig. 2). The reduced scattering coefficient spectra from the initial IAD output matched well with Mie theory, besides at the prominent water absorption peaks near 1500 and 2000 nm, where coefficients were overestimated. After fitting (Eq. (4)) and rerunning IAD, the resulting reduced scattering coefficient spectra matched well with Mie theory [39] (Fig. 2(A)), with a median coefficient difference of 0.06 mm^-1 and 0.32 mm^-1 for the low (0.54% v/v) and high (4.78% v/v) concentration solutions respectively. The final absorption spectra also aligned well with literature values for water absorption [42] (Fig. 2(B)), with a median coefficient difference of 0.08 mm^-1. Lastly, the anisotropy spectrum had a median difference of 0.03 compared to Mie theory [39] (Fig. 2(C)). Notably, the initially assumed constant anisotropy for calculating reduced scattering and absorption coefficients had a relatively small effect on the final anisotropy. For example, an assumed anisotropy of zero (g = 0), which is ∼0.55-0.93 different from the actual value across the spectrum, only produced a median error of 0.065 (Fig. S1A-C). Furthermore, the assumed constant refractive index (n = 1.33) and constant calibration standard reflectance (R = 99%), which are known to vary from n = 1.34-1.28 [42] and R = 99.5–92.5% (Spectralon) respectively across the spectrum, had minimal effect on the resulting determined optical properties (Fig. S1D-I).

 

Fig. 2. (A) Mie theory (red), measured (black) and fit (blue) reduced scattering coefficient spectra; (B) mean (± standard deviation) measured (black) and literature (red | Hale and Querry) absorption coefficient spectra; and (C) Mie theory (red) and measured (black) anisotropy factor spectra of 0.54% and 4.78% v/v 1.1-µm polystyrene microspheres in water solutions.

Download Full Size | PDF

3.2 Ex vivo tissue optical properties

The optical properties of fat, muscle, tendon, and nerve tissues from five rats and five human cadavers were determined using the measurement and processing protocol validated in section 3.1. The initial IAD output of the reduced scattering coefficient (Fig. 3) decreased with wavelength in all tissues, except at the locations of prominent absorption peaks from hemoglobin and water, due to cross-talk between calculated reduced scattering and absorption coefficients. These were corrected using the two-stage IAD approach protocol (Eq. (4)) for all tissues besides tendon, which had much higher scattering than all other tissues and did not follow a power law decay. For the rat (Fig. 3(A)), nerve maintained a significantly greater reduced scattering coefficient than muscle and fat across almost the entirety of the wavelength range, and significantly lower reduced scattering coefficients compared to tendon (Fig. S2A). Similarly, the reduced scattering coefficients from human nerve (Fig. 3(B)) were significantly higher (Fig. S2B) than muscle and fat in the visible (VIS | 352-680 nm) and near-infrared (NIR | 680-1000 nm) and were lower compared to tendon. The scattering anisotropy of rat muscle, fat, and nerve was additionally determined (Eq. (6)) to accurately simulate diffuse reflectance in section 3.3. Each tissue maintained anisotropies greater than 0.85 across the NIR and SWIR (1000-2500 nm) (Fig. 3(A) inset). In the VIS, the anisotropy rapidly decreased with shorter wavelengths, corresponding to increased Rayleigh scattering.

 

Fig. 3. Mean (± standard error) reduced scattering coefficient spectra for fat (blue), muscle (red), tendon (green), and nerve (black) from (A) rats and (B) human cadavers. Inset in (A) displays anisotropy factor spectra for rat nerve, muscle, and fat. Mean (± standard error) absorption coefficient spectra for fat (blue), muscle (red), tendon (green), and nerve (black) from (C) rats and (D) human cadavers. See Dataset 1 [63] and Dataset 2 [64] for underlying data.

Download Full Size | PDF

The corrected absorption coefficient spectra from the two-stage IAD protocol are shown in Fig. 3(C)-(D). The rat (Fig. 3(C)) and human cadaver (Fig. 3(D)) tissues exhibited four well-known tissue chromophores, namely: hemoglobin (blood), lipid, collagen, and water. These chromophores and their absorption peaks are well characterized and can be found in multiple reviews [38,43]. Hemoglobin dominated the VIS for each tissue type. The rat tissues exhibited absorption mainly from deoxyhemoglobin (433 and 556 nm peaks) in the nerve and muscle spectra, oxyhemoglobin (414, 543, and 577 nm peaks) in tendon, and mixed oxygenation in the fat spectra. Rat nerve absorption was significantly different from that of muscle in portions of the VIS, but not fat and tendon across most of the VIS (Fig S2C). Absorption from blood was greater in human cadaver nerve, tendon, and muscle compared to the rat, and all human tissues exhibited oxyhemoglobin absorption. In the NIR, there was a significant decrease in muscle absorption, which continued into the SWIR until ∼1340 nm. Rat fat had a lower mean absorption in the NIR compared to nerve and tendon, whose absorption coefficients nearly matched in this region. For human tissue, the mean NIR absorption of nerve was not statistically different from that of fat and muscle (Fig. S2D), with tendon having the highest absorption in the NIR. Water absorption peaks at 970, 1180, 1456, and 1936 nm were present in each tissue type for rats and humans. Additionally, there were prominent lipid absorption peaks at 1210, 1730, and 1760 nm in fat spectra and, to a lesser degree, in nerve. Fat also had distinct lipid absorption peaks at 920, 2312, and 2352 nm.

3.3 Identification of a promising spectral region for embedded nerve detection

The effective penetration depth was calculated from the absorption and reduced scattering coefficients (Eq. (7)) of rat (Fig. 4(A)) and human tissues (Fig. 4(B)) to identify an optimal spectral region for embedded nerve detection. The effective penetration depth was highest between ∼700-1350 nm. For rat tissues, muscle had the greatest effective penetration depth, reaching over 15 mm at 1084 nm. This was followed by fat, which maintained around 1.35 mm of effective penetration depth across most of the NIR, reaching a maximum of ∼1.40 mm at 1084 nm. The effective penetration depth was around 810 µm for nerve in the NIR reaching a maximum ∼878 µm at 1084 nm. For tendon the maximum effective penetration depth was ∼526 µm at 1116 nm. Muscle again had the highest penetration depth in the NIR for the cadaver tissues. However, greater penetration depths were achieved for fat in the VIS and much of the SWIR, due to its relatively lower blood and water absorption respectively. Tendon had the shortest penetration depth in the VIS and NIR due to its high scattering. Most importantly, the SWIR wavelengths near 1100 nm maximized photon penetration depth in all rat and human tissues.

 

Fig. 4. Optical property-based identification of a promising spectral region for embedded nerve detection. Mean (± standard error) effective penetration depth for (A) rat and (B) human fat (blue), muscle (red), tendon (green) and nerve (black). Monte Carlo simulated diffuse reflectance (± standard error) in the wavelength region of highest effective photon penetration depth for (C) rat and (D) human tissues (2 mm of fat (blue), 3 mm of muscle (red), 1 mm of tendon (green), and 1 mm of nerve (black)). PC1 (black) and PC2/PC3 (red) coefficients as a function of wavelength from (E) rat and (F) human simulated diffuse reflectance show a local maximum coefficient of variance near 1210 nm for PC2 and PC3 respectively.

Download Full Size | PDF

The spectral region with the greatest photon penetration depths (∼700-1350 nm) was further analyzed to identify a sub-region with optimal contrast. Monte Carlo simulations were performed to calculate diffuse reflectance spectra between 700-1350 nm for the rat (Fig. 4(C)) and human (Fig. 4(D)) tissues. Tendon had higher mean reflectance and muscle had lower mean reflectance compared to nerve across the entire spectral region. For the rat, fat exhibited similar reflectance values to nerve but distinct spectral dissimilarities near 1210 nm due to lipid absorption. PCA was employed to identify characteristic spectral features of nerve in this spectral window. PC1 contributed to 97.2% of variance, however, only PC2 (1.97%) scores completely separated rat nerve from surrounding tissue spectra. PC1 loadings seemed to directly follow differences in reduced scattering between the tissues, whereas PC2 identified a maximum positive coefficient at ∼1210 nm, aligning with lipid absorption (Fig. 4(E)). The results from human tissues were similar (Fig. 4(F)). PC1 (88.9% of the variance) loadings again trended with scattering, with PC1 scores for nerve significantly different from the other tissues (p < 0.01). PC3 (2.48% of the variance) scores also separated nerve from fat and muscle with the PC3 coefficient spectrum reaching a maximum near 1210 nm. The differences in total reflectance between nerve, muscle, and tendon, as well as the distinct spectral features near 1210 nm separating nerve from fat presented a promising spectral region for providing nerve contrast at maximal photon penetration. With this optimal spectral region established, the next step was identifying specific wavelengths that provided the best label-free nerve contrast for real-time nerve imaging implementation.

3.4 In vivo shortwave infrared spectral imaging of nerves

To identify specific wavelength combinations optimal for in vivo nerve imaging, a SWIR hyperspectral camera covering the optimal wavelength regime determined from optical properties and simulation, near 1210 nm, was selected. Spectral images (1000-1700 nm) were acquired from 19 in vivo rat sciatic nerves with surrounding muscle, fat, and skin. Since tendon was not present in the approved rat sciatic nerve preparation, ex vivo mouse tail tendons were imaged separately for reference. Diffuse reflectance spectra, normalized to a 99% reflectance puck, were averaged for manually segmented regions of nerve, muscle, fat, and tendon (Fig. 5(A)). The mean spectrum for each nerve and the mean spectrum from all surrounding tissues in each image were fed into PCA to identify optimal wavelengths for nerve contrast. Like the Monte Carlo results, PC2 provided the best separation of nerve from surrounding tissue types (Fig. 5(B)), even though it only accounted for 8% of the overall variance. PC2 loadings displayed maximal negative coefficients at 1100 and 1455 nm, zero coefficients at 1145 and 1380 nm, and a maximal positive coefficient at ∼1190 nm (Fig. 5(C)). The 1455 nm wavelength contributed most to the variance and aligned with water absorption. Taking a ratio between the 1455 nm image and its nearest zero coefficient at 1380 nm (R_1455/1380) provided high contrast of nerve (Fig. S3). However, since the presence of water can be highly variable during surgical procedures and is prohibitive of embedded nerve imaging due to significant absorption, wavelengths less dominated by water were considered for further analysis. Ratioing images at the local maximum (1190 nm) and minimum (1100 nm) coefficients (R_1190/1100) provided sufficient nerve contrast (Fig. 5(D)) while maintaining the potential for embedded nerve detection. The mean (± standard deviation) R_1190/1100 of nerve was 0.51 ± 0.06, whereas fat was 0.45 ± 0.05, muscle 0.41 ± 0.03 and ex vivo tendon 0.58 ± 0.02 (Fig. 5(E)). ANOVA and a Tukey-Kramer multiple comparison test confirmed that nerve had a significantly different ratio value compared to the other tissues (p < 0.05 in all cases).

 

Fig. 5. Shortwave infrared imaging enhances rat nerve contrast. (A) Mean (± standard error) diffuse reflectance spectra from regions of rat fat (blue | n = 12), muscle (red | n = 19) and nerve (black | n = 19) extracted from hyperspectral cubes from left and right in vivo rat sciatic nerve preparations. Ex vivo mouse tendon spectra (green | n = 6) is shown for reference. (B) Box and whisker plots depicting the distribution of scores for principal components (PCs) 1-5 of nerve spectra (black) and surrounding tissue spectra (red), with PC2 being the most distinct between nerve and surrounding tissue. (C) PC2 coefficients show high variance (1100 nm, 1090 nm and 1455 nm) and low variance (1145 nm and 1380 nm) wavelengths for enhancing nerve contrast. (D) Ratiometric images of 1190 nm to 1100 nm provide enhanced nerve contrast (0.4–0.7 scale | arrows point to nerves). The inset depicts the peak 1085 nm reflectance on a 0–1 scale. (E) R_1190/1100 values of nerve (± standard deviation) are significantly different than surrounding tissue (* p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001).

Download Full Size | PDF

To determine the depth at which nerve could be visualized under layers of tissue, slices of fat and muscle were placed over excised nerves and SWIR spectral images were collected. The experiment was repeated for three different rat nerves. R_1190/1100 images maintained visually discernable nerve contrast until around 1 mm of overlayed fat (Fig. 6(A)). The nerve detection depth, where the NC fell to 1/e of the original (uncovered) nerve contrast, was 0.68 ± 0.17 mm for overlayed fat (Fig. 6(B)). This study was repeated with muscle, where the nerve detection depth was calculated to be 0.60 ± 0.11 mm (Fig. 6(C)).

 

Fig. 6. Detecting nerves under fat and muscle tissue. (A) Ratiometric images (R_{1190nm/1100nm}) of two nerves, normalized to the uncovered nerve (0 mm) ratiometric signal, maintain nerve contrast with 0-1000 µm of overlayed fat. (B) Mean (± standard deviation) nerve contrast (NC) drops to 1/e at 0.68 ± 0.17 mm of overlayed fat. (C) Mean (± standard deviation) NC drops to 1/e at 0.60 ± 0.11 mm of overlayed muscle. The mean and standard deviation of the NC decay curve and the mean R² value is given by least squares exponential fits from the experimental repetitions (n = 3 rat nerves).

Download Full Size | PDF

4. Discussion

Iatrogenic nerve injuries (INJs) can be life-altering for a patient and present significant medicolegal consequence. To avoid INJs, multiple optical techniques are being explored and have demonstrated enhanced nerve visualization and detection. To further advance the efficacy of optical nerve detection, the primary goal of this study was to address the lack of reports providing fully characterized nerve optical properties and their relation to surrounding tissue. Optical properties can be utilized to identify mechanisms of optical nerve contrast and allow for the simulation of photon propagation. A Monte Carlo simulation performed with these properties can provide diffuse reflectance for any irradiation-detection geometry and can assist in optimizing nerve contrast from surrounding tissues using iterative approaches. This application can be further extended to fluorescence-based buried nerve detection, which can be optimized using optical properties and a known fluorophore spectrum and concentration [44,45]. Moreover, any diffuse optical technique (Raman spectroscopy, polarimetry, spatial frequency domain imaging, optical coherence tomography, etc.) can exploit known nerve optical properties along with their corresponding optical interaction parameters for the simulation and optimization of nerve detection [46]. With this in mind, the optical properties of rat and human nerve, muscle, fat, and tendon were determined. But first, a measurement and postprocessing protocol was validated using a well-characterized standard to instill confidence in the measured optical properties.

Scattering agents such as intralipid [47,48] and titanium dioxide (TiO₂) [49,50] are often used as reduced scattering coefficient validation standards. However, each requires careful characterization of particle size distributions to be accurately modeled with Mie Theory. Furthermore, the recipe for these agents can change over time and vary between manufacturers, requiring recharacterization. To avoid these issues, off-the-shelf polystyrene microspheres were chosen due to their narrow particle size distributions and small number of ingredients, namely microspheres in water. This standard allowed for simple and accurate calculation of theoretical scattering properties, which may improve consistency in optical property calculations across studies [51,52]. The microsphere solution’s reduced scattering coefficient, determined with the integrating sphere-IAD approach, aligned well with that given by Mie Theory across the VIS to SWIR, except at the highly absorbing bands of water (Fig. 2). The recently proposed two-stage IAD method [36,37] accurately corrected the overestimations in reduced scattering coefficients, indicating its robustness. An additional measurement of unscattered transmittance allowed for calculating the anisotropy factor using Beer’s Law. The shorter pathlength used for this measurement facilitated adequate signal-to-noise while producing anisotropy factors that aligned with Mie Theory. This method may be beneficial in cases where thin samples are needed for sufficient unscattered light detection, but thicker samples are necessary for diffuse reflectance and transmittance. Overall, the chosen particle size of 1.1 µm produced a reduced scattering coefficient decay shape and a range of anisotropy factor values like that of tissue. To validate absorption, water was chosen because it is well characterized. The calculated absorption coefficients from the dual-stage IAD algorithm matched well with those in the literature (Fig. 2). Like the microspheres, the use of water avoids variations based on the manufacturer, which can be problematic with other chromophores and dyes. Although measurement of IR properties allowed for validation using water, studies limited to the VIS regime would require the addition of a VIS chromophore to the solution, which could cause aggregation of the microspheres. Overall, the simple and robust method of microspheres-in-water validation provided confidence in the measurement and processing procedures, which were then directly applied to characterize nerve and surrounding tissue.

The optical properties of nerve were distinct from that of surrounding tissues. Nerve’s reduced scattering coefficient was higher than that of fat and muscle across the VIS and NIR for rat and human tissues (Fig. 3), like the results reported by Wisotsky et al. [33]. The dual-stage IAD approach effectively removed cross-talk in the reduced scattering coefficient spectra of nerve, muscle, and fat. However, tendon, which had significantly higher scattering than the other tissues, did not fit the Mie-Rayleigh power law, thereby prohibiting the use of dual-stage IAD for this tissue. Future studies will be necessary to identify and validate a function that accurately fits tendons reduced scattering spectrum, which may consist of adding decay phases to the Mie-Rayleigh power law. Regardless, tendon did not have observable cross-talk until the highly absorbing water bands, likely due to its high scattering compared to absorption across most of the broadband (Fig. 3). Notably other studies characterizing tendon optical properties have been limited to a few wavelengths or restricted to one spectral region and overall report widely varying property coefficients, animal types and optical property measurement techniques [53–56]. Our study provides a broadband characterization and a direct comparison to other tissue types, but further studies will be necessary to provide a reason for tendon’s distinct reduced scattering decay shape, which is likely due to a distinct scatterer size distribution. When comparing rat to human reduced scattering spectra, there was minimal change for tendon and muscle. However, the human nerve and fat had much lower scattering in the IR and across the VIS-IR, respectively, compared to rat. The likely explanation would be damage to lipid structures due to freezing of the human tissues since lipid accounts for 60-94% of white adipose tissue composition [57] and 50.2% of nerve dry weight [58]. Due to these potential effects, the scattering anisotropy was not additionally characterized for the human cadaver tissues. However, the anisotropy factor was determined for rat nerve, muscle, and fat. The anisotropy was between 0.85 and 0.95 for each tissue across most of the broadband (Fig. 3(A) inset), similar to what has been reported for soft tissues [38]. When approaching the VIS from the IR, the anisotropy had a consistent decay, with similar values and spectral shape as those reported for porcine skin [59]. Regarding the absorption coefficient, differences between nerve and surrounding tissues could be explained by the distinct spectral contributions of tissue chromophores hemoglobin, lipid, and water. Rat and human tissues had similar absorption curves for each tissue type respectively (Fig. 3(C), (D)). However, one difference was the higher blood absorption in human tissues compared to rat tissues, which could be due to the human tissues having more vascular density compared to tissues of smaller mammals like the rat. Human fat was the only exception, having low overall blood and water absorption compared to that of rat fat. It should be noted that although care was taken to ensure adequate tissue hydration and minimal tissue intrinsic blood (hemoglobin) loss, concentrations of these chromophores in the surgical field can be highly variable during a procedure, negating this assumption. As such, direct in vivo optical properties measurements would be valuable for assessing coefficient variance within bloody surgical fields. Regardless, lipid absorption and SWIR scattering, which had low optical property variance and are less affected by blood absorption, provided the contrast that was exploited in the latter half of the study. The most distinct absorptive dissimilarity from nerve was the strong lipid absorption in fat near 1210 nm and 1730 nm (Fig. 3). Overall, nerve had higher blood absorption, lower lipid absorption and higher water absorption than fat. This is corroborated by the single other report comparing nerve and fat IR absorption coefficients, which backed these findings with histological validation [32].

Having determined the optical properties of nerve and surrounding tissues, we sought to utilize these insights to detect nerves buried in tissue. Calculating the effective penetration depth revealed that wavelengths from ∼700-1350 nm achieved maximal photon penetration in each tissue (Fig. 4(A), (B)). Monte Carlo simulation diffuse reflectance in this spectral region along with PCA confirmed optimal nerve contrast near 1210 nm, which aligned with lipid absorption (Fig. 4(E), (F)). To verify these results in vivo and identify specific wavelengths for optimal nerve contrast, data was collected from a preclinical rat sciatic nerve model using a SWIR hyperspectral camera covering 1000-1700 nm. PCA identified multiple wavelengths contributing to nerve contrast, corresponding to lipid (1100 and 1190 nm) and water absorption (1455 nm) (Fig. 5(C)). These results were similar to those of Schols et al., who used probe-based DRS from 350-1830 nm to differentiate nerve from fat using reflectance at wavelengths 965 nm, 1180 nm, 1210 nm, and the max reflectance between 1050 -1210 nm [25]. We further confirmed that nerve’s spectroscopic contrast could be visualized with R_1190/1100 and R_1455/1380 ratiometric imaging (Fig. 5(D)). Marginal nerve contrast could also be achieved using the max reflectance in the spectrum at 1085 nm (Fig. 5(D) inset). However, false positives could be observed from each tissue type (fat, muscle, and skin) in various 1085 nm images. The ratiometric approach removed most of these false positives, however, a few were still observed. For example, the transition edge from skin to muscle produced R_1190/1100 ratios like nerve, which was a downside to using a ratio with an intermediate value between skin and muscle or any other tissue type. The R_1455/1380 images avoided this issue because nerve had a lower ratio value than all other tissue types and was unaffected by tissue type transitions. As such, collecting images at more than two wavelengths could be used to improve nerve contrast at the cost of adding filters (e.g., filter wheel) and sequential acquisition. Considering the 10-15 ms exposure times utilized in this study, sequential image acquisition at a few wavelengths could be accomplished at video-rate. Future human studies will be necessary to quantify nerve detection accuracy for both surgeon and computer, from video-rate feedback, using a set number of wavelengths. Lastly, some false positives/negatives were due to specular reflections; however, this could be easily mitigated by using a cross-polarized setup, namely illuminating and detecting with orthogonal linear polarization states.

Notably, to our knowledge, this is the first study employing label-free spectral SWIR imaging for nerve visualization. Previous label-free nerve imaging studies have used VIS diffuse reflectance or VIS polarized light imaging to gain nerve contrast [31,60]. These techniques offer a low-cost and non-invasive method for visualizing nerves but are limited to detecting superficial nerves under <300-400 µm of overlayed tissue [60]. SWIR ratiometric imaging improves the depth of nerve detection to ≥600 µm of overlayed fat and muscle (Fig. 6), which could help avoid injury to an occluded nerve. Apart from label-free techniques, NIR exogenous fluorescence from contrast agents is also being employed for nerve imaging [28,29]. For example, a recent study achieved nerve contrast under 1-3 mm of mouse fat and muscle using ∼700 nm fluorescence from oxazine derivatives [29]. Interestingly, based on the optical properties results in our study, nerve visualization depth could be maximized by developing nerve-specific SWIR fluorophores near 1100 nm [61,62]. Although labeled approaches provide increased depths of nerve visualization, label-free SWIR imaging presents a promising method in cases where contrast agents cannot be utilized (e.g., patient allergies, drug availability and cost etc.). However, further studies will be necessary to validate label-free SWIR imaging for intraoperative human nerve visualization.

5. Conclusion

In this study, the optical properties of rat and human nerve, muscle, fat, and tendon were determined from 352-2500 nm to inform and advance optical techniques and systems employed for nerve sensing and imaging. These properties facilitated the identification of a promising spectral region in the SWIR that provided both nerve contrast and maximal photon penetration depth. In vivo SWIR spectral imaging of a rat sciatic nerve preparation was performed for further validation, where optimal nerve contrast was achieved using 1190/1100 nm ratiometric imaging. Nerve contrast was maintained with ≥600 µm of overlayed fat and muscle. Overall, the results provide valuable insights for optimizing the optical contrast of nerves, including those embedded in tissue, which could lead to improved surgical guidance and nerve-sparing outcomes.