2.1.

Dosimeter System

A diagram and photographs of the probe-based experimental dosimeter are provided in Fig. 1. The system housed five illumination sources: four temperature-controlled laser diodes (405, 639, 735, and 785 nm, WorldStarTech, Toronto, Ontario) and an Ocean Optics HL-2000 Tungsten Halogen white light source (Ocean Optics, Dunedin, Florida). These sources were coupled to a motorized $6\times 1$ fiber switch (DiConFiberoptics Inc., Richmond, California) through 200-μm fibers to enable rapid switching between channels. The output of the fiber switch was coupled to the patient’s skin through an Ocean Optics R200-7-UV/VIS optical fiber probe that consisted of seven 200-μm fibers in the arrangement shown in Fig. 1(b). During measurements, the probe was held in contact with the patient’s skin and light was delivered to the tissue through the central channel, shown in blue in Fig. 1(b). The six fibers surrounding the illumination fiber transmitted the remitted light through a motorized six-position filter wheel (ThorLabs FW102C, Newtown, New Jersey) and into an Ocean Optics USB4000 spectrometer with a spectral range between 346 and 1038 nm. The lasers, filter wheel, and spectrometer were USB-ready devices, while the DiCon fiber switch and white light source’s internal shutter were controlled with an NI USB-6501 DAQ board. All control connections were consolidated in a powered USB hub connected to a laptop. The system was operated using a simple software interface written in Labview (National Instruments, Austin, Texas).

Fig. 1

(a) Schematic of the multichannel dosimeter setup showing system components. (b) Fiber arrangement of the patient-interface (probe tip), which is used in light compression contact-mode. The source fiber is represented in blue. The surrounding six fibers transmit light remitted from the tissue through the filter wheel and into the spectrometer. Photographs of the internal optical assembly and the chassis are shown in (c) and (d), respectively.

The PDT dosimetry measurements in this study were acquired with the white light source, 405 nm (blue channel) and 639 nm (red channel) laser diodes, and thus, the 735- and 785-nm laser diode channels were not used. Excitation filtering for fluorescence measurements was accomplished using a 1-mm-thick 435-nm color glass long-pass filter (Schott, Mainz) for blue channel excitation and a 650-nm long-pass interference filter (ThorLabs) for red excitation. The filters were selected to reduce, but not eliminate, the excitation signal to facilitate simultaneous measurement of the excitation and emission intensities in the same spectrum. While this was readily achievable with red channel excitation, the high attenuation of blue light in the presence of blood often reduced the blue light excitation signal to undetectable levels, even with low filtering efficiency.

Low-cost 635-nm laser diodes typically operate 2 to 4 nm above 635 nm, as was the case with the laser diode used in this system, so this was actually at 639 nm. Since the absorbance peak of PpIX falls off precipitously within this range, these lasers are rather inefficient at exciting PpIX fluorescence, a reality that results in relatively high minimum detectable PpIX concentrations. This is addressable by integrating 635-nm laser diodes with higher-wavelength tolerances at significantly higher cost.

Measurements were automatically acquired in the following sequence: (1) white light reflectance with no filtering, (2) red channel fluorescence, (3) blue channel fluorescence. To ensure that measured spectra were within the dynamic range of the spectrometer, measurement exposure times were optimized automatically during the acquisition. This was achieved by acquiring a spectrum at an initial exposure time of 200 ms, checking that the measured signal was within a specified range and then either confirming the measurement was valid or automatically adjusting the acquisition time and repeating the measurement. The minimum and maximum times allowed were typically set at 10 and 1000 ms, respectively. Most fluorescence measurements were made at 300 or 50 ms, respectively, for blue and red channels, while white light measurements usually required 1 s. After each measurement, the fiber switch was positioned to a dark channel and a second spectrum acquired to allow real-time background subtraction of the ambient room light. Thus, a total of six spectra were recorded for an acquisition sequence of all three channels. The total acquisition time for this sequence was typically 25 s, limited primarily by the overhead in changing filter wheels and the long exposure time for the white light reflectance measurements. All measured spectra were background subtracted and corrected for exposure time before being analyzed using the protocols described below.

2.2.

White Light Reflectance Spectroscopy Data Processing

Analyses of white light spectra were used for two purposes: (1) to correct the fluorescence measurements for the effects of optical propagation in tissue and (2) to estimate endogenous physiological parameters, such as whole blood volume fraction, oxygen saturation, and blood vessel density. The analysis was accomplished by applying a Monte Carlo–based look-up-table (LUT) approach similar to what has been reported previously^28,29 to relate measured light intensity to the wavelength-dependent optical properties within the sampled volume, including the absorption coefficient (${\mu }_a$) and the reduced scattering coefficient (${\mu }_s^{\prime }$). In this study, LUT was generated for broadband light using validated Monte Carlo code³⁷ with the geometry configured to mimic the six discrete detector fibers interspaced around a central source fiber, as in Fig. 1(b). All fiber diameters were 0.2 mm and the source to detector spacing was set to 0.25 mm. Simulations assumed a homogenous medium with 20 values for ${\mu }_s^{\prime }$, covering the range [0.1 to 50] ${\mathrm{mm}}^{-1}$, and 13 values for ${\mu }_a$, over the range of [0 to 10] ${\mathrm{mm}}^{-1}$, resulting in 260 unique optical property combinations. Other simulation parameters include a Henyey-Greenstein phase function with anisotropy of $\mathrm{g}=0.9$, index of refraction of the $\text{medium}=1.37$ and $\text{fibers}=1.45$, and the numerical aperture of source and collection $\text{fibers}=0.22$. Each simulation was initialized with ${10}^7$ photons and returned the collected reflectance intensity.

Measured spectra were calibrated by relating the wavelength-dependent collected light intensity measured in a reference phantom composed of 2% Intralipid ($I_{\mathrm{ref}}^{\text{meas}}$) to the LUT-based intensity returned for the known optical properties³⁸ of the phantom ($R_{\mathrm{calib}}^{\mathrm{LUT}}$). Thus, the calibration was given as

This calibration converts collected light intensity measured in a sample ($I_{\text{sample}}^{\mathrm{meas}}$) in units of counts/millisecond to reflectance ($R_{\text{sample}}^{\mathrm{meas}}$) expressed in units of the percentage of launched photons that were collected from the Monte Carlo LUT.

After calibrating using Eq. (1), reflectance spectra were fit to an empirical relationship describing the effects of optical absorption and scatter on the remitted spectrum, a process that facilitated estimation of these parameters in the measured volume. Background scattering was estimated with a power-law function as

Here, the absorption coefficients of blood and melanin are given by ${\mu }_a^{\text{blood}}$ and ${\mu }_a^{\mathrm{melanin}}$, respectively. The absorption due to whole blood can be described by

For measurements performed on skin in the presence of melanin ($R_{\text{sample}}^{\mathrm{skin},\mathrm{mel}}$), the attenuation attributable to the absorption of melanin was estimated using a Beer-Lambert based factor as

Prior to fitting, spectra were smoothed into discrete bins 5 pixels in width (equal to 1.5 nm), with a mean and standard deviation calculated within each bin. The smoothed reflectance spectra were fit over the [480 to 800] nm wavelength range, with light intensity $<480\mathrm{nm}$ not included due to poor system transmission and sensitivity at those wavelengths. For measurements of skin, the algorithm estimated BVF, Sat, $a$, $b$, and $c_{\mathrm{MEL}}$; $r_v$ was unable to be fit as a free parameter, but was instead assumed to be 20 μm. For measurements of optical phantoms, the algorithm estimated BVF, Sat, $a$, and $b$, and did not include either melanin or $C_{\mathrm{ves}}$. The spectral fitting algorithm was coded in MATLAB^® using the lsqnonlin subroutine and the standard deviation of each binned pixel was used as a weighting factor. The algorithm also returned confidence intervals for fitted parameters as previously described⁴¹ using nlparci subroutine. The qualities of spectral fits were quantified using a reduced chi-squared metric, ${\chi }^2=\frac{1}{v}\sum [{(R_{\text{data}}-R_{\text{model}})}^2/{\sigma }^2]$.

2.3.

Fluorescence Spectral Analysis

Quantifying fluorescence activity from the measured spectra generally followed a three-step process: (1) the raw spectra (after background subtraction and exposure time correction) were calibrated to the reference phantom, (2) these spectra were then corrected for the influence of optical properties using results from the white light spectroscopy measurements, and (3) a linear least squares spectral fitting routine decoupled the specific PpIX fluorescence from other fluorescent signals, such as autofluorescence and photoproducts. The details of this process are described here.

Collected fluorescence intensities $J_{\text{sample}}^{\mathrm{meas}}({\lambda }_x)$ were initially calibrated using

Next, the calibrated fluorescence spectrum, ${\mathrm{FL}}_{\text{sample}}^{\mathrm{meas}}({\lambda }_x)$, is corrected for the influence of background optical properties by using an empirical correction factor based on sampled reflectance intensities, which is an approach described extensively,^31,42–44 and is given as

Once corrected for optical properties, ${\mathrm{FL}}_{\text{sample}}^{\mathrm{corr}}({\lambda }_x)$ spectra were analyzed as a linear combination of basis spectra representing tissue autofluorescence, PpIX fluorescence, and PpIX photoproduct (PP) fluorescence. For blue excitation, the emission was fit by

Prior to fitting, the calibrated fluorescence spectra were reduced into discrete bins 5 pixels in width (equal to 1.5 nm), with a mean and standard deviation calculated within each bin. The fitted waveband for ${\mathrm{FL}}_{\text{sample}}^{\mathrm{corr}}({\lambda }_{blue})$ was [600 to 800] nm and that for ${\mathrm{FL}}_{\text{sample}}^{\mathrm{corr}}({\lambda }_{\text{red}})$ was [675 to 800] nm. The fitting algorithm was coded in a MATLAB^® script and used the lsqnonlin subroutine to perform a Levenberg-Marquardt least squares fitting regime that minimized the error between model predictions and measured data, with each pixel weighted by the standard deviation. The fitting procedure returned estimates and confidence intervals of [$c_{\text{auto}}^{\text{blue}},c_{\mathrm{PpIX}}^{\text{blu}e},c_{\mathrm{PP}1}^{\text{blue}},c_{\mathrm{PP}2}^{\text{blue}}$] for blue and [$c_{\text{auto}}^{\text{red}},c_{\mathrm{PpIX}}^{\text{red}},c_{b1}^{\text{red}},c_{b2}^{\text{red}},c_{b3}^{\text{red}}$] for red excitation.

To show the increased accuracy enabled by spectral fitting, we also examined the simple integral of the corrected fluorescence spectra ${\text{Area}}_{\mathrm{PpIX}}^{\text{source}}$ over a waveband of the remission spectra, essentially bypassing the spectral fitting algorithm (the final step in the analysis process). The trapz subroutine in MATLAB^® was used to numerically integrate ${\mathrm{FL}}_{\text{sample}}^{\mathrm{corr}}({\lambda }_{\text{blue}})$ over [620 to 720] nm, and ${\mathrm{FL}}_{\text{sample}}^{\mathrm{corr}}({\lambda }_{\text{red}})$ over [680 to 720] nm, providing estimates of ${\mathrm{Area}}_{\mathrm{PpIX}}^{\text{blue}}$ and ${\mathrm{Area}}_{\mathrm{PpIX}}^{\text{red}}$, respectively.

2.4.

System Validation Using Tissue-Simulating Phantoms

Phantom experiments were used to achieve four objectives: (1) characterize the linearity of response to PpIX concentration for both fluorescence channels, (2) determine the minimum detectable concentration of PpIX for both fluorescence channels, (3) determine the accuracy of the white light reflectance analysis in estimating scattering parameters and chromophore concentrations, and (4) examine the stability of the fluorescence measurements to background optical properties with and without correction. Phantoms were constructed using Intralipid (20% Frenius-Kabi, Bad Homburg, Germany), bovine whole blood (7200811 Lampire Biological Inc., Pipersville, Pennsylvania), and PpIX (P8293, Sigma-Aldrich, St. Louis, Missouri). All phantoms were composed of 5% Tween20 (P1379, Sigma-Aldrich) to mitigate aggregation. To explore linearity of response and minimum detection limits of PpIX fluorescence, measurements were acquired in phantoms composed of 1% Intralipid, and 1% BVF, and over a range of $[\mathrm{PpIX}]=[0.122,0.244,0.488,0.977,1.95,3.90,7.80,15.6,31.3,62.5,125,250,500,1000,2000,4000]\mathrm{nM}$. To examine accuracy of white light reflectance analysis and stability to changes in optical properties, measurements were acquired in phantoms prepared using volume fractions of Intralipid [1, 2, 3] % and whole blood [0.5, 1, 2, 3] %. This resulted in 12 phantom combinations, each of which was sampled using serial dilutions of PpIX to achieve $[\mathrm{PpIX}]=[15.6,62.5,250,1000,4000]\mathrm{nM}$. For each phantom composition, white light reflectance, and blue- and red-excited fluorescence measurements were acquired, and measurements were repeated three times. To determine the lower limit of detection for each fluorescence metric ($c_{\mathrm{PpIX}}^{\text{blue}},c_{\mathrm{PpIX}}^{\text{red}},{\text{Area}}_{\mathrm{PpIX}}^{\text{blue}},{\text{Area}}_{\mathrm{PpIX}}^{\text{red}}$), the data were fit to a line forced through zero, and the lower limit was identified as the highest [PpIX] at which the metric deviated $>25\%$ from the fit. Estimated optical parameters were compared with known phantom properties (i.e., BVF and ${\mu }_s^{\prime }$) using the Pearson correlation coefficient ($r$) to show linearity, and the mean residual was calculated as $\overline{\mathrm{resid}}=100[(\text{estimated}-\text{known})/\text{estimated}]$.

2.5.

Pilot clinical Study

A clinical feasibility study was conducted to establish the feasibility of using the device, to determine initial metrics of system performance, and to acquire information about interpatient variability. The study was approved by the institutional review board, and 19 patients with AK, identified in a standard clinical dermatology practice (MS Chapman, Dartmouth Hitchcock Medical Center), were enrolled. Informed consent was obtained in writing from patients prior to initiation of treatment. PDT procedure began with topical application of ALA (Levulan Kerastick, DUSA Pharmaceuticals, Wilmington, Massachusetts) to the lesion area. Patients were instructed to wait for 1 h in the outpatient waiting room, with care taken to avoid exposure to direct sunlight. Patients were treated with a noncoherent blue light source (BLU-U Blue Light Photodynamnic Therapy Illuminator, DUSA Pharmaceuticals) with the prescribed exposure of 10 J/cm² (8 to 15 min) for each patient. Immediately following treatment, patients were asked to assess their maximal pain level experienced during treatment using the visual analog scale (VAS) from 0 to 10; this was recorded as an assessment of treatment response in this study. Optical measurements were performed at three time points during the treatment protocol: (1) prior to administration of ALA, (2) following 1-h incubation with ALA and before light illumination, and (3) immediately following administration of the treatment light dose. All measurements were recorded with the probe tip in gentle contact with the surface of the AK lesion. Fluorescence measurements prior to ALA administration were used to record patient-specific basis spectra for the tissue autofluorescence and were used in the spectral fitting algorithms described in Eqs. (8) and (9).

Optical spectra recorded were checked for integrity using the fit quality metric ${\chi }^2$. Fluorescence measurements from patients prior to ALA administration were used to determine a patient-specific basis spectra for the tissue autofluorescence. Analysis of optical spectra from patients provided information about PpIX generation and PDT dose as well as vascular changes in response to PDT.

2.6.

PDT Dose Metrics for Pilot Clinical Data

Optical measurements of PpIX fluorescence were used to estimate metrics of the therapeutic dose delivered during treatment. Assuming that the dose can be characterized by observed changes to PpIX fluorescence (and that adequate oxygen was either supplied by perfusion or diffusion through the skin), the dose is proportional to the amount of PpIX consumed through photobleaching during treatment, as shown previously.²⁵ The absolute photodynamic dose was calculated as

The relationship between the two fluorescence channels was also examined by calculating a ratio-based metric of PpIX fluorescence intensity sampled by the red and blue channels as

Patient-reported pain was used as a preliminary response metric in this pilot study. Metrics of PDT dose and patient-reported pain were compared statistically using a Kruskal–Wallis one-way analysis of variance, with significance associated with $p<0.05$.

2.7.

Depth Sensitivity Analysis Using Monte Carlo Simulations

In order to characterize the origin depth of PpIX fluorescence sampled using the blue and red channels, we performed a set of Monte Carlo simulations. The simulation geometry approximated the optical property distribution in the skin as a seven-layer structure, as described previously in Ref. 40, and considered three cases with varying spatial distributions of PpIX within the tissue. Case 1 included homogenous PpIX throughout all of the tissue volume with ${\mu }_{\mathrm{af}}^{\mathrm{PpIX}}$ selected to approximate $[\mathrm{PpIX}]=1000\mathrm{nM}$. Case 2 considered superficially distributed PpIX, with the absorption of PpIX at deeper depths ($Z<300\mu \mathrm{m}$) decreased by a factor of 7, while the PpIX at shallow depths was unchanged. Case 3 considered deeply distributed PpIX, with the absorption of PpIX at shallow depths ($Z>300\mu \mathrm{m}$) decreased by a factor of 7, while the PpIX at deeper depths was unchanged. The simulations were performed using validated Monte Carlo code;⁴⁶ each simulation initialized ${10}^7$ photons. The collected fluorescence intensity for the measurement and depth of origin for each collected fluorescence photon were recorded and used to estimate the 80% sampling depth ($Z_{\mathrm{FL}}^{80\%}$) using the quantile subroutine in MATLAB^®.

3.1.

Tissue Phantom Measurements

3.1.1.

Linearity of fluorescence response

To characterize the response of the dosimeter system to changes in PpIX concentration ([PpIX]) and to establish the limits of detection, measurements were acquired in tissue phantoms composed of Intralipid 1% and whole blood 1% and [PpIX] $\in$ [0.1 to 4000] nM. Figure 2(a) shows a representative fluorescence spectrum resulting from blue light excitation for a phantom containing $[\mathrm{PpIX}]=500\mathrm{nM}$, and the corresponding model fit. The plot shows the fitted contributions of background tissue autofluorescence, PpIX, and the very small contributions from photoproducts to the total fluorescence. $c_{\mathrm{PpIX}}^{\text{blue}}$ versus [PpIX] data are shown on linear [Fig. 2(b)] and log [Fig. 2(c)] scales to illustrate the quality of the linear relationship over a wide range of [PpIX] values. The log-scale plot in Fig. 2(c) allows identification of the lower limit of detection of $c_{\mathrm{PpIX}}^{\text{blue}}$ at 1.95 nM, below which the $c_{\mathrm{PpIX}}^{\text{blue}}$ versus [PpIX] relationship deviates from linearity. For comparison, Fig. 2(d) shows the integrated metric of fluorescence (${\text{Area}}_{\mathrm{PpIX}}^{\text{blue}}$), which does not utilize spectral fitting and results in an elevated lower limit of detection of 250 nM. This substantial difference in sensitivity is due to the fitting algorithm adequately decoupling fluorescence from PpIX and background fluorophores, while the integrated metric is confounded by these factors at low [PpIX] values.

Fig. 2

Fluorescence measurements in protoporphyrin IX (PpIX) dilution series as measured with blue excitation light. Phantoms were constructed with Intralipid 1%, whole blood 1%, and Tween 5%. (a) A representative spectral fit is shown for $[\mathrm{PpIX}]=500\mathrm{nM}$. Sixteen dilutions were sampled between 4000 nm and 0.1 nM. The fitted PpIX fluorescence versus known PpIX concentration is shown (b) on a linear scale and (c) on a log scale with the lower limit of detection identified at 1.95 nM. The integrated fluorescence area versus PpIX concentration is shown in (d) with the lower limit of detection identified at 250 nM.

The data in Fig. 3 show the fluorescence response of the same phantom set as in Fig. 2 measured with the red excitation laser. Figure 3(a) shows representative measured and fitted spectra for $[\mathrm{PpIX}]=1000\mathrm{nM}$. The data in Figs. 3(b) and 3(c) show a linear response for $c_{\mathrm{PpIX}}^{\text{red}}$ versus [PpIX] from [250 to 4000] nM and Fig. 3(d) shows the ${\text{Area}}_{\mathrm{PpIX}}^{\text{red}}$ versus [PpIX]. Lower limits of PpIX detection for the red channel were identified for both spectral fitting (250 nM) and integrated fluorescence (1000 nM). The sensitivity of the system to red channel fluorescence is far lower than for blue light excitation, which is attributable to the absorption coefficient of PpIX being $\sim 30$-fold higher at 405 nm than at 639 nm.

Fig. 3

Fluorescence measurements in PpIX dilution series as measured with red excitation light. Phantoms were constructed with Intralipid 1%, whole blood 1%, and Tween 5%. (a) A representative spectral fit is shown for $[\mathrm{PpIX}]=1000\mathrm{nM}$. Sixteen dilutions were sampled between 4000 nm and 0.1 nM. The fitted PpIX fluorescence versus known PpIX concentration is shown (b) on a linear scale and (c) on a log scale with the lower limit of detection identified at 250 nM. The integrated fluorescence area versus PpIX concentration is shown in (d) with the lower limit of detection identified at 1000 nM.

3.1.2.

White light spectroscopy

Optical measurements were performed in 12 sets of tissue-simulating phantoms with different amounts of Intralipid (volume $\text{fraction}=$ [1 to 3]%) and whole blood ($\mathrm{BVF}=$ [0.5 to 3]%) to yield biologically relevant scattering and absorption properties. Figure 4(a) shows an example of a measured white light spectrum sampled in phantom with 1% Intralipid and 1% BVF. The black circles indicate measured data points and the corresponding model fit is shown in the red line. Differences between data and model fits are shown in the residual displayed below the spectrum. Figure 4(b) shows the estimated versus known ${\mu }_s^{\prime }$ at two wavelengths [440, 639] nm. The scattering estimates showed a linear relationship versus known values in all phantoms ($\text{slope}=1.00$, $r=0.99$), with a mean residual between model estimates and the linear fit of $\overline{\mathrm{resid}}({\mu }_s^{\prime })<3\%$. Figure 4(c) shows the estimated versus known BVF with a linear relationship ($\text{slope}=0.98$, $r=0.98$) over the range $\mathrm{BVF}=$ [0.5 to 3]% and mean residual $\overline{\mathrm{resid}}(\mathrm{BVF})<15\%$.

Fig. 4

(a) Representative white light spectra sampled in tissue phantoms for Intralipid 1% and blood volume of 1%. Estimated versus known values for (b) reduced scattering coefficient at blue and red excitation wavelengths and (c) blood volume fraction. Error bars on (b) and (c) represent variation between phantoms.

3.1.3.

Fluorescence correction for optical property distortion

Optical measurements of the 12 different combinations of absorption and scattering were used to examine the influence of optical properties on estimates of PpIX fluorescence and the capacity to correct for these effects. Figure 5(a) shows the variation of raw PpIX fluorescence estimated from ${\mathrm{FL}}_{\text{sample}}^{\mathrm{uncorr}}({\lambda }_{\text{blue}})$ among the 12 phantoms. Here residual error is expressed as the percentage difference between the PpIX fluorescence estimated in each phantom and the average of all phantoms; each phantom contains the same amount of [PpIX], i.e., 1000 nM. These data show that the raw fluorescence is substantially distorted by variations in background scattering and absorption, with $\overline{\mathrm{resid}}<37\%$. Figure 5(b) shows residual error for corrected PpIX fluorescence, estimated from ${\mathrm{FL}}_{\text{sample}}^{\mathrm{corr}}({\lambda }_{\text{blue}})$ as given in Eq. (5). Comparison of Figs. 5(a) and 5(b) shows a reduction in error, with $\overline{\mathrm{resid}}$ dropping to $<9\%$ for the corrected estimates of PpIX. Figures 5(c) and 5(d) show uncorrected and corrected PpIX fluorescence estimated from the red channel, with an improvement in $\overline{\mathrm{resid}}$ values from $<20$ to $<8\%$ provided by the optical property correction factor. Inspection of the error plots do not show a trend with either Intralipid or whole blood, indicating that the resulting corrected estimates of PpIX are insensitive to variations in both scattering and absorption.

Fig. 5

(a) and (b) The residual error between fitted PpIX fluorescence emission spectra excited by blue excitation light for uncorrected and corrected spectra, respectively. (c) and (d) The residual error between fitted PpIX fluorescence emission spectra excited by red excitation light. Residual is calculated as the percentage deviation for fitted PpIX fluorescence normalized to average of estimates from all phantoms.

3.2.

Pilot Clinical Data

3.2.1.

Clinical PDT observations

Of the 19 patients enrolled in the study, 15 were used in this analysis; two were excluded due to an instrument error (IDs 10 and 11), and one patient was excluded due to a sampled BVF that exceeded the absorption bounds of the LUT (ID 2). Additionally, data from one patient showed erroneous features in the spectra (ID 8), which were identified by a metric of fit quality (${\chi }^2>3$) and are likely due to confounding environmental influences (e.g., sufficient coupling to skin surface, influence of room lights, movement artifacts, etc.). Figure 6 shows pain assessed on the VAS immediately following PDT treatment, with 12 of 15 patients reporting nonzero pain; $\overline{\text{pain}}=4.8+/-2.3$, $\text{range}=$ [0 to 9]. While the number of patients is small, the reported pain approximates a normal distribution and spans the full VAS range.

Fig. 6

Visual analog scale (VAS) pain as reported by patients immediately after photodynamic therapy (PDT) treatment. # indicates that pain was not assessed on ID1.

Figure 7 shows representative fluorescence and white light spectra measured from the AK lesion on patient ID 1. Fluorescence spectra sampled from the lesion using both blue and red excitation sources show strong PpIX fluorescence emission prior to treatment light delivery [Figs. 7(a) and 7(c), respectively], while post-treatment fluorescence activity was much reduced [Figs. 7(b) and 7(d)]. The associated photobleaching rates were ${\mathrm{PB}}_{\mathrm{PpIX}}^{\text{blue}}=97\%$ and ${\mathrm{PB}}_{\mathrm{PpIX}}^{\text{red}}=100\%$. White light reflectance spectra presented in Figs. 7(e) and 7(f) show differences in vascular physiology between pre- and post-Tx, with increases in BVF and Sat observed. The analysis presented in Fig. 7 is representative of the spectral fits performed on most of the patients enrolled in the study.

Fig. 7

Representative optical measurements of fluorescence emission with excitation with (top) blue, (middle) red, and (bottom) white light reflectance from patient ID1. Left-hand panels show measurements 1 h following aminolevulinic acid administration (pre-Tx). Right-hand panels show measurements immediately after administration of treatment light (post-Tx).

Figure 8(a) shows PpIX fluorescence excited by blue light measured before and after light treatment for each patient. These data show substantial interpatient variability in the accumulation of PpIX after the 1-h incubation time; ${\overline{c}}_{\mathrm{PpIX}}^{\text{blue}}=3.4+/-3.1$, $\text{range}=$ [0 to 9.3]. For patients with nonzero PpIX fluorescence, the majority of PpIX was consumed during treatment with high observed photobleaching rates ${\overline{\mathrm{PB}}}_{\mathrm{PpIX}}^{\text{blue}}=89+/-11\%$, $\text{range}=$ [62 to 100] %, and thus, the absolute dose (${\mathrm{Dose}}_{\mathrm{PpIX}}^{\text{blue}}$) closely followed the pre-Tx PpIX fluorescence, with ${\overline{\mathrm{Dose}}}_{\mathrm{PpIX}}^{\text{blue}}=24.7+/-22.5$. When patient-reported pain after treatment was grouped into high ($\mathrm{VAS}>=5$) and low ($\mathrm{VAS}<5$) pain, both $c_{\mathrm{PpIX}}^{\text{blue}}$ and ${\mathrm{Dose}}_{\mathrm{PpIX}}^{\text{blue}}$ were predictive of treatment-induced pain (with $p<0.01$ for both), with the latter shown in Fig. 8(b). Figure 8(c) shows that the photobleaching expressed as a percentage of pre-Tx (${\mathrm{PB}}_{\mathrm{PpIX}}^{\text{blue}}$) did not show a significant indication for patient-reported pain ($p<0.24$); it should be noted that this analysis of photobleaching included zero values for undetectable PpIX.

Fig. 8

PpIX fluorescence measured in response to blue excitation. (a) Patient-specific fitted PpIX fluorescence pre- and post-Tx. Box-plots show comparison of (b) absolute PDT dose and (c) percent PpIX photobleaching with patient reporting low pain ($\mathrm{VAS}<5$) and high pain ($\mathrm{VAS}>=5$).

A similar analysis was completed for the red channel fluorescence measurements, as shown in Fig. 9. These data also show substantial interpatient variability in pre-Tx PpIX accumulation of PpIX(${\overline{c}}_{\mathrm{PpIX}}^{\text{red}}=2.9+/-2.8$, $\text{range}=$ [0 to 9.1]) and substantial photobleaching (${\overline{\mathrm{PB}}}_{\mathrm{PpIX}}^{\text{blue}}=8.9+/-17\%$, $\text{range}=$ [46 to 100] %). While less significant than for the blue channel measurements, both $c_{\mathrm{PpIX}}^{\text{red}}$ and ${\mathrm{Dose}}_{\mathrm{PpIX}}^{\text{red}}$ showed a significant difference for patients reporting low and high pain ($p<0.03$ and $p<0.04$), which was also not observed for the calculated photobleaching values, ${\mathrm{PB}}_{\mathrm{PpIX}}^{\text{red}}$ ($p<0.09$).

Fig. 9

PpIX fluorescence measured in response to red excitation. (a) Patient-specific fitted PpIX fluorescence pre- and post-Tx. Box-plots show comparison of (b) absolute PDT dose and (c) percent PpIX photobleaching with patient reporting low pain ($\mathrm{VAS}<5$) and high pain ($\mathrm{VAS}>=5$).

Analysis of white light spectra returned quantitative descriptions of tissue composition and vascular physiology. Figure 10 shows optical tissue parameters for all patients before and after light treatment. The observed average values of these parameters for pretreatment measurements and the change in values after treatment are blood volume fraction $\overline{\mathrm{BVF}}=0.9+/-1.1$ (%) and $\overline{\mathrm{\Delta }\mathrm{BVF}}=0.0+/-1.0$ (%) [Fig. 10(a)]; microvascular saturation $\overline{\mathrm{SAT}}=42+/-36$ (%) and $\overline{\mathrm{\Delta }\mathrm{SAT}}=16+/-48$ (%) [Fig. 10(b)]; reduced scattering coefficient ${\overline{\mu }}_s^{\prime }(639\mathrm{nm})=2.1+/-0.3{\mathrm{mm}}^{-1}$ and ${\mathrm{\Delta }\overline{\mu }}_s^{\prime }(639\mathrm{nm})=0.1+/-0.3{\mathrm{mm}}^{-1}$ [Fig. 10(c)]; and melanin fraction in epidermis ${\overline{c}}_{\mathrm{MEL}}=14+/-3$ (%) and ${\overline{\mathrm{\Delta }c}}_{\mathrm{MEL}}=1.6+/-4.6$ (%) [Fig. 10(d)]. Comparing changes in physiological parameters with PDT dose metrics showed a moderate inverse correlation between the change in BVF and pain ($r=-0.55\mathrm{m}$, $p<0.08$), and a moderate positive correlation between ${\mathrm{Dose}}_{\mathrm{PpIX}}^{\text{blue}}$ and the change in hemoglobin saturation ($r=0.44$, $p<0.08$); those correlations were not significant in this small study.

Fig. 10

Patient-specific optical parameters measured before and after PpIX PDT treatment: (a) blood volume fraction, (b) microvascular saturation, (c) reduced scattering coefficient at 637 nm, and (d) volume fraction of melanocytes in the epidermis.

3.2.2.

Depth sensitivity of analysis of two-color PpIX measurement

PpIX fluorescence intensities sampled by both blue and red excitation on each lesion were highly correlated ($r=0.90$, $p<0.001$). A ratiometric comparison of red to blue PpIX fluorescence (${\text{Ratio}}_{\mathrm{RB}}$) showed substantial interpatient variability (${\overline{\text{Ratio}}}_{\mathrm{RB}}=0.92+/-0.51$, $\text{range}=$ [0.23 to 2.23]), as shown in Fig. 11(a). It should be noted that ${\text{Ratio}}_{\mathrm{RB}}$ was calculated only for patients showing detectable PpIX before treatment, excluding three additional patients from this analysis. In principle, ${\text{Ratio}}_{\mathrm{RB}}$ should reveal coarse information on the depth of the PpIX distribution, with larger ${\text{Ratio}}_{\mathrm{RB}}$ values associated with PpIX fluorescence originating from deeper depths within the skin tissue; however, this has not been shown explicitly. Interestingly, ${\text{Ratio}}_{\mathrm{RB}}$ was an indicator of pain, with elevated ${\text{Ratio}}_{\mathrm{RB}}$ values significantly higher for patients reporting lower pain; this is shown in Fig. 11(b).

Fig. 11

Patient-specific ratio of red to blue PpIX fluorescence; * marks indicate zero PpIX. Box-plot shows comparison of ratio with patient reporting low pain ($\mathrm{VAS}<5$) and high pain ($\mathrm{VAS}>=5$).

A series of simulations were performed to characterize the sensitivity of two-color fluorescence measurements of PpIX to the depth-dependent distribution of PpIX in the skin. The simulation results are summarized in Fig. 12. Case 1 simulates homogenously distributed PpIX, with red channel sampling PpIX from $3+$-fold deeper than the blue and the ratio of red to blue fluorescence intensities normalized to 1. Case 2 considers superficially distributed PpIX; here, both channels sample superficial PpIX. The decrease in deeper PpIX resulted in a disproportionate decrease in intensity sampled with red compared with blue (with 74 and 10% reductions in red and blue intensities), such that the resulting ${\text{Ratio}}_{\mathrm{RB}}$ decreased to 0.3. Case 3 considers deeply distributed PpIX; here, the signal sampled from both channels now originates from deeper layers. The decrease in superficial PpIX saw intensity sampled with blue decreased much more than with red (with 80 and 15% reductions in blue and red intensities), yielding an increase of ${\text{Ratio}}_{\mathrm{RB}}$ to 4.4. These data indicate that the ratiometric analysis of PpIX fluorescence sampled by red and blue channels can provide a qualitative description of the PpIX distribution, with larger values of ${\text{Ratio}}_{\mathrm{RB}}$ associated with more deeply distributed PpIX and, conversely, smaller values of ${\text{Ratio}}_{\mathrm{RB}}$ associated with superficially distributed PpIX.

Fig. 12

Monte Carlo simulation data showing influence of PpIX depth distribution on the sampled depth of fluorescence origin and the ratio of collected light intensities. Three distributions were considered: (a) uniform PpIX throughout, (b) superficial PpIX, and (c) deep PpIX. The 80% fluorescence sampling depth is reported for blue and red excitation channels, and the red/blue ratio is given for each case.