2.1.

Research Population Recruitment and Tissue Handling

2.1.4.

Sample collection

Kidneys were collected during surgical nephrectomies. There was no deviation from routine clinical surgical practice. Specimens collected during surgery were handled under the supervision of a pathologist prior to, during, and after the optical property measurements. If the surgeon decided that any excised kidney required immediate evaluation by pathology, that specimen was not included in this study.

Nephrectomy specimens were brought directly from the operating room to the laboratory for optical measurements. A pathologist performed the initial dissection of the specimen as per standard surgical pathology grossing procedures on a bench adjacent to the diffuse reflectance spectroscopy setup. This ensured a clean, flat-cut surface against which the spectroscopy probe could be positioned with good contact. By visual inspection, the pathologist then identified normal and neoplastic tissue on which optical measurements would be performed. A representative dissected kidney is shown in Fig. 1. The measurements were taken using a custom diffuse reflectance spectroscopy system. The spectroscopy probe, shown schematically in Fig. 2, made gentle contact with the surface of the tissue being interrogated. The nephrectomy specimen was then immediately taken to the surgical pathology laboratory, where it was accessioned and processed as per routine practice.

Fig. 1

Photograph of a sectioned kidney containing a renal cell carcinoma (RCC).

Fig. 2

Schematic of the optical probe used in contact with the tissue surface for diffuse reflectance measurements. The source fiber launched broadband light into the tissue, and 10 detection fibers, located at distances of 1 to 10 mm from the source fiber, collected the diffusely reflected signal.

2.2.

Diffuse Reflectance Spectroscopy

White light, steady-state diffuse reflectance measurements were performed at each of the normal and neoplastic regions of the specimen using techniques that have been developed and optimized in the laboratory over the past decade. Our instrumentation and method of data reduction for performing quantitative diffuse reflectance spectroscopy have been described in detail elsewhere.^13,14 Broadband light from a xenon lamp was delivered to the surface of the tissue via an optical fiber with a 400-μm core diameter, which is terminated in a probe of our own design and construction (Fig. 2). The probe also holds a linear array of 20 detection fibers (200-μm core diameter) positioned at regular 1-mm spacings at distances of 1 to 20 mm from the source fiber. In the studies reported here, adequate signal-to-noise was obtained from the first 10 of these fibers, corresponding to source-detector separations ranging from 1 to 10 mm. We confirmed that measurements from these 10 fibers were sufficient by systematically deleting data from the most remote fiber and refitting the remaining points until a change in the returned optical properties was observed. For all kidneys analyzed, five to eight source-detector separations were adequate for accurate recovery of optical properties.

The flat, polished surface of the probe made good contact with the clean, flat-cut surface of the tissue under investigation, allowing white light from the source fiber to be injected into the specimen surface. The distal ends of the detection fibers were gathered into a bundle that terminated in a ferrule. At the end of the ferrule, the fibers were arranged in a row and positioned in the focal plane of a 0.275-m imaging spectrograph (SpectraPro-275, Princeton Instruments/Acton, Acton, MA). The fibers were imaged on to the surface of a thermoelectrically cooled, $512\times 512$, 16-bit CCD camera (Pixis512, Princeton Instruments/Acton, Trenton, NJ). The detector fiber position was imaged along one axis of the CCD array, and the spectral information over the range of 630 to 800 nm was recorded along the other. Thus, at each wavelength in the data set, the experiment captured a spatially resolved attenuation measurement. Typical signal integration times for measurements described here were approximately 5 s. The raw data were corrected for system throughput and fiber cross-talk and then processed through a Savitzky-Golay smoothing filter. From these attenuation measurements, the absorption (${\mu }_a$) and transport scattering coefficients (${\mu }_s^{\prime }$) of the tissue were extracted.

2.3.

Extraction of Optical Properties

A broadly accepted theoretical description of light propagation in biological tissues is the Boltzmann transport equation.¹⁵ This equation treats light propagation as energy transport, in which absorption and scattering are the dominant attenuation mechanisms. We have previously demonstrated the accuracy of an approximation to this expression known as the $P_3$ approximation, which expands the Boltzmann transport equation in Legendre polynomials up to third order.¹⁶ The diffuse reflectance predicted by this model explicitly depends on the optical properties of the sample and can therefore be used to extract these properties from measured data. The $P_3$, rather than the diffusion approximation, was used here because of its validity over a wider range of possible optical properties.

When light is introduced into tissue, it is attenuated by wavelength-dependent scattering and absorption. In a diffuse reflectance measurement, a fraction of the multiply scattered light is emitted from the tissue surface, where it is collected by one of the probe fibers, as illustrated schematically in Fig. 2. With increasing distance from the source fiber, the amplitude of the reflectance signal is diminished, because light exiting the tissue at larger source-detector separations has traversed a greater path length. These spatially resolved diffuse reflectance data may be fit using an appropriate theoretical model to obtain ${\mu }_a$ and ${\mu }_s^{\prime }$ without assumptions regarding the composition of the tissue. Here, the $P_3$ model of light transport was fit to the measured reflectance using a constrained nonlinear optimization routine found in the MATLAB optimization toolbox (Mathworks, Natick, MA). The fitting function, ${\chi }^2$, to be minimized was the sum of the squared error between the measured dataset at each wavelength and the $P_3$ model, scaled by the measured diffuse reflectance at each source-detector separation, as shown in the equation

2.4.

Monte Carlo Model of Interstitial PDT

PDT of renal tumors would be performed by introducing an optical fiber or fibers directly into the tumor under image guidance. Simulations of light propagation in tissue from cylindrical diffusing tip fibers were performed using Monte Carlo (MC) modeling. The particular code used was developed in our laboratory and incorporates a rigorous model of cylindrical diffusing fibers implemented on a graphics processing unit.²⁰ Briefly, the model uses a realistic description of cylindrical diffusing fibers to launch photon packets into the surrounding tissue, with standard MC techniques being used to control photon propagation.²¹ Interactions between photon packets and the diffuser after launch are also considered.

MC simulations were used to model two scenarios. The first consisted of an ellipsoidal tumor embedded in a normal kidney. The tumor had equatorial radii of 1.1 cm and a polar radius of 1.3 cm, with a semi-infinite region of normal kidney tissue surrounding it. This tumor size was determined by averaging measured tumor dimensions from clinical images and assuming an ellipsoidal tumor shape. In this case, the optical properties used were those extracted from $P_3$ fits to measured diffuse reflectance in both normal kidney and tumor tissue, with each being assigned to the appropriate region of the sample. Based on the treatment wavelengths used for currently available and anticipated PDT photosensitizers, simulations were run at 630, 652, 670, 692, and 780 nm for each kidney sample.

The second model consisted of a homogeneous sample of either normal kidney or tumor tissue. Optical properties were set uniformly to match those extracted from $P_3$ fits to measured diffuse reflectance in normal or tumor tissue. Simulations of both normal and tumor tissue were run at 50 discrete wavelengths from 630 to 800 nm for each kidney sample. This allowed for modeling of a characteristic light propagation distance at each of the measured wavelengths.

For all MC simulations, 1,000,000 photons were launched from a 2-cm-long diffuser. In all cases, the tissue sample was represented with a $100\times 100\times 100$ grid of cuboid voxels, each with dimensions of 0.026 cm. A Henyey-Greenstein phase function with g$=0.85$ was used for all simulations.

3.1.

Diffuse Reflectance Spectroscopy

We analyzed data from 14 kidneys. The results obtained from two kidneys were not interpretable due to low signal levels. Figure 3 shows representative diffuse reflectance spectra acquired from the RCC region of a freshly excised human kidney. The spectra were corrected for measured system throughput, background signal, and cross-talk in the CCD. Individual traces correspond to different source-detector separations on the diffuse reflectance probe, which range from approximately 1 to 10 mm.

Fig. 3

Measured diffuse reflectance spectra at the first five source-detector separations collected from a kidney region identified as RCC. Spectra were corrected for measured optical system throughput, background, and cross-talk in the CCD. Individual traces correspond to different source-detector separations on the diffuse reflectance probe, with their locations relative to the source fiber given in the legend.

3.2.

Kidney Optical Properties

The optical properties of tumor and normal kidney tissue were extracted at each wavelength by fitting a $P_3$ model of light transport to the measured data using a constrained nonlinear optimization routine as described in Sec. 2. An example of such a fit is shown in Fig. 4. The individual data points are corrected, normalized diffuse reflectance measurements made at 780 nm in a representative RCC. Each point was acquired at a different source-detector separation [r (mm)] using the probe as described above. The fit shown in Fig. 4 yielded optical properties of ${\mu }_a=0.072{\mathrm{mm}}^{-1}$ and ${\mu }_s^{\prime }=1.40{\mathrm{mm}}^{-1}$.

Fig. 4

The individual data points are corrected, normalized diffuse reflectance measurements made at 780 nm in a representative RCC. Each point was acquired at a different source-detector separation [r (mm)]. Error bars are the standard deviations of three repeated measurements of the same sample. The fit of the $P_3$ model to these data yielded optical properties of ${\mu }_a=0.072{\mathrm{mm}}^{-1}$ and ${\mu }_s^{\prime }=1.40{\mathrm{mm}}^{-1}$.

By analyzing spatially resolved reflectance at each measured wavelength, absorption and scattering spectra over the measured spectral range were determined for each kidney sample. Examples from two lesion types are shown in Fig. 5. In these two cases, ${\mu }_a$ was lower in the tumor than in the normal kidney tissue. However, this trend was not reproducible across all of the kidney samples (data not shown). The spectra exhibit decreased absorption and scattering with increasing wavelength, as expected. The absorption feature near 759 nm corresponds to a peak characteristic of deoxyhemoglobin.

Fig. 5

Values of (a) ${\mu }_a$ and (b) ${\mu }_s^{\prime }$ extracted from fitting the $P_3$ model to measured diffuse reflectance spectra from a kidney diagnosed with RCC. In this case, both ${\mu }_a$ and ${\mu }_s^{\prime }$ are lower for tumor tissue than for kidney parenchyma. However, this was not reproducible for all RCC samples. (c) ${\mu }_a$ and (d) ${\mu }_s^{\prime }$ extracted from fitting the $P_3$ model to measured diffuse reflectance spectra from a urothelial carcinoma and adjacent normal parenchyma. In this case, ${\mu }_s^{\prime }$ is higher in the tumor.

The recovered values of ${\mu }_a$ ranged from 0.016 to $0.14{\mathrm{mm}}^{-1}$ for RCC conventional; 0.003 to $0.22{\mathrm{mm}}^{-1}$ for oncocytoma; 0.02 to $0.062{\mathrm{mm}}^{-1}$ for urothelial carcinoma; 0.14 to $0.31{\mathrm{mm}}^{-1}$ for RCC, papillary type 1; 0.5 to $0.75{\mathrm{mm}}^{-1}$ for RCC, papillary multifocal; and 0.017 to $0.13{\mathrm{mm}}^{-1}$ for kidney parenchyma. The recovered values of ${\mu }_s^{\prime }$ ranged from 0.48 to $1.92{\mathrm{mm}}^{-1}$ for RCC conventional; 1.23 to $2.5{\mathrm{mm}}^{-1}$ for oncocytoma; 1.06 to $1.33{\mathrm{mm}}^{-1}$ for urothelial carcinoma; 0.25 to $0.3{\mathrm{mm}}^{-1}$ for RCC, papillary type 1; 1.32 to $1.83{\mathrm{mm}}^{-1}$ for RCC, papillary multifocal; and 0.11 to $5.51{\mathrm{mm}}^{-1}$ for kidney parenchyma. These recovered optical properties yielded values of the transport albedo [$a^{\prime }\equiv {\mu }_s^{\prime }/({\mu }_a+{\mu }_s^{\prime })$] that ranged from 0.49 to 0.99. The recovery of transport albedos lower than 0.98 indicates that the $P_3$ approximation was an appropriate choice over the diffusion approximation.¹⁶

3.3.

Monte Carlo Modeling

Monte Carlo simulations of light propagation from cylindrical diffusing fibers were run using the optical properties extracted from each of the kidney samples. A representative fluence distribution generated by the delivery of $100\mathrm{J}{\mathrm{cm}}^{-1}$ of 630-nm light from a 2-cm-long diffuser is shown in Fig. 6. This wavelength corresponds to the long wavelength absorption maximum of the FDA-approved photosensitizer Photofrin. The figure shows fluence deposited in the surrounding tissue using optical properties that were derived from a kidney diagnosed with urothelial carcinoma. In order to examine this fluence distribution more carefully, cuts were made through the fluence distribution perpendicular to the diffuser axis. The results of this are shown in Fig. 7(a) for a 2-cm-long diffuser delivering $100\mathrm{J}{\mathrm{cm}}^{-1}$, with each contour line corresponding to a fluence increment of $50\mathrm{J}{\mathrm{cm}}^{-2}$. The outer black line in the figure represents the boundary between tumor and normal tissue, as described previously. The inner black line represents the boundary between tumor tissue and the diffuser.

Fig. 6

3D rendering of the simulated fluence distribution around an optical fiber with the proximal end of a 2-cm-long diffusing segment of the fiber positioned at $z=0.3\mathrm{cm}$ above the boundary separating the tumor and normal tissue regions. Tumor optical properties were set to ${\mu }_a=0.062{\mathrm{mm}}^{-1}$, ${\mu }_s^{\prime }=1.34{\mathrm{mm}}^{-1}$, and $g=0.85$, which correspond to values extracted from an urothelial carcinoma at 630 nm, as shown in Fig. 5(c) and 5(d). Voxel size was $0.026\times 0.026\times 0.026{\mathrm{cm}}^3$.

Fig. 7

Contour plots of cuts through simulated fluence distributions around a 2-cm-long diffusing fiber at (a) 630 nm and (b) 780 nm, with a total fluence of 100 J cm⁻¹ delivered. Each contour line corresponds to a fluence increment of 50 J cm⁻². The simulated volume consists of an ellipsoidal tumor region, with equatorial radii of 1.1 cm and a polar radius of 1.3 cm, surrounded by a semi-infinite region of normal tissue. The boundary between tumor and normal tissue is shown as the outer black line; the boundary with the diffuser is shown as the inner black line. The optical properties correspond to values extracted from the tumor and normal regions of a kidney diagnosed with urothelial carcinoma, as shown in Fig. 5(c) and 5(d).

Figure 5 reproduces in the human kidney the well-known observation that absorption and scattering in tissue both decrease with increasing wavelength in this spectral range. It is therefore appealing to consider the use of photosensitizers with longer wavelength absorption in interstitial PDT of the kidney. This potential is shown in Fig. 7, where fluence deposition is modeled at 630 and 780 nm. At the longer wavelength, the penetration of light is greatly improved compared to 630 nm. For a delivered fluence of $100\mathrm{J}{\mathrm{cm}}^{-1}$, $\sim 20\%$ of the tumor volume received a light dose of at least $50\mathrm{J}{\mathrm{cm}}^{-2}$ at 630 nm. This increased to approximately 50% of the tumor volume at 780 nm. Additionally, nearly 30% of the tumor received a light dose of $100\mathrm{J}{\mathrm{cm}}^{-2}$ or more at 780 nm, compared to $\sim 11\%$ at 630 nm. This trend was reproducible across all samples, with the percentage of the tumor receiving a prescribed light dose increasing by an average of 2.2-fold as a result of changing the wavelength from 630 to 780 nm. Increasing the fluence delivered by the cylindrical diffuser to $600\mathrm{J}{\mathrm{cm}}^{-1}$ extended the $50\mathrm{J}{\mathrm{cm}}^{-2}$ contour to the simulated boundary of the tumor (not shown).

In order to describe a characteristic distance of light propagation from a cylindrical diffusing fiber, MC simulations were run using homogeneous optical properties in a semi-infinite tissue volume surrounding a 2-cm-long diffuser. A zeroth order modified Bessel function of the second kind, $K_0$, is a useful analytic approximation to light emitted from an infinitely long cylinder. Thus, a function of the form

Fig. 8

Results of fitting a zeroth order modified Bessel function of the second kind, $K_0$ [Eq. (2)], to a cut through the simulated fluence around a 2-cm-long diffuser. The fluence plotted is the radially averaged simulated fluence at the axial midpoint of the diffuser. Optical properties were set to ${\mu }_a=0.017{\mathrm{mm}}^{-1}$, ${\mu }_s^{\prime }=1.55{\mathrm{mm}}^{-1}$, and $g=0.85$, which correspond to values extracted from a representative RCC at 780 nm. From such fits, the characteristic propagation distance, $\delta$, is obtained.

Fig. 9

Simulated light propagation distance, $\delta$, based on Bessel function fit to the simulated radial distribution of fluence in Fig. 8, for a kidney diagnosed with RCC.

Table 1

Mean characteristic light propagation distances, in mm, at five selected wavelengths, separated by diagnosis. Wavelengths were selected to correspond with the absorption peaks of Photofrin (630 nm), Foscan (652 nm), methylene blue (670 nm), benzoporphyrin derivative (BPD, 692 nm), and next-generation photosensitizers (780 nm). Uncertainties shown are standard deviations over multiple samples, and values in parentheses indicate range over multiple samples.