2.1.

Instrumentation

An LED-based, digital micromirror device (Texas Instruments) projection system (Fig. 1) was used to provide both structured illumination at discrete wavelengths (630, 730, and 850 nm) for the determination of optical properties and provide planar illumination for fluorescence over a 30×40 mm field of view. A 12 bit CCD camera (Lumenera) is used to capture reflectance and fluorescence signals from the tissue. For SFDI, a cross polarization scheme blocks specular reflections. For fluorescence, the analyzer is replaced with a long-pass filter with a cut-off at 650 nm. The image plane of the CCD is matched to the same field of view of the projection. For SFDI measurements, six spatial frequencies (0 to 0.5 mm⁻¹), are projected at three spatially shifted phases (0, 120, and 240 deg) per wavelength. Total acquisition times take approximately 4 s, but are allowed to vary in order to utilize the full dynamic range of the camera. A preset exposure of 2 s is used to acquire fluorescence images as a practical measure, though the full dynamic range of the camera was not exploited.

Fig. 1

Instrument diagram.

2.2.

Model

While there are a variety of models available to render quantitative fluorescence information (e.g., Durkin ¹⁰), we decided to explore the model developed by Gardner for reasons that will be discussed.¹¹ This model is an empirical approach for planar excitation fluorescence that accounts for the presence of tissue absorption and reduced scattering based off of semi-infinite, homogeneous Monte Carlo simulations. Assuming planar illumination, this model is simplified and reduced to a one-dimensional (1D) problem. Since SFDI is also based on planar illumination, these two models are easily integrated and can be applied to each image pixel. SFDI provides the μ_a and μ_s ^′ at both excitation and emission wavelengths as inputs to Gardner's model that use these to correct the generation and existence of the fluorescence. In this model, the fluorescence correction factor, X _1D, is determined by the following:

The empirical coefficients, C _n and k _n are defined in Table 1 of the Gardner paper;¹¹ however, they are dependent on the diffuse reflectance, R _d. Since R _d is a function of μ _a, μ _s ^′ and index of refraction, it can be calculated from the values determined by SFDI using a Monte Carlo simulation. δ refers to optical penetration depth, which can be calculated directly: [TeX:] $\delta (\lambda) = {1 \mathord{/ {\vphantom {1 {\sqrt {3\mu _a \left(\lambda \right)\left[ {\mu _a \left(\lambda \right) + \mu _s ^\prime \left(\lambda \right)} \right]} }}} \kern-\nulldelimiterspace} {\sqrt {3\mu _a \left(\lambda \right)[ {\mu _a \left(\lambda \right) + \mu _s ^\prime \left(\lambda \right)}]} }}$ $\delta \left(\lambda \right)=1/\sqrt{3{\mu }_a\left(\lambda \right)\left[{\mu }_a\left(\lambda \right)+{\mu }_s^{\prime }\left(\lambda \right)\right]}$ . In the validation study, coefficients were determined under the condition where n _sample/n _air = 1.33. In the in vivo study, an index condition of n _sample/n _air = 1.38 was employed.

For the determination of the optical properties at the excitation wavelength (μ_{a _ex}, μ_{s _ex} ^′), the SFDI results calculated at 630 nm can be plugged directly into this equation [Eq. 1]. The fluorescence emission detected by this instrument, however, is integrated across all wavelengths in the spectral emission band. To properly account for this, the emission spectrum-weighted average of both absorption and scattering would have to be determined. With a long pass filter at 650 nm in the system, we consider our signal detects emissions spanning 650 to 750 nm. Assuming that melanin, oxy-, and deoxy hemoglobin are the primary chromophores in skin tissue, we use the absorption results from all three wavelengths to determine chromophore concentration and subsequently estimate the average absorption value (〈μ_a〉_em) within the 650 to 750 nm range. To estimate the average reduced scattering (〈μ_s ^′〉_em) in that range, we assume that it can be interpolated from the three measured wavelengths, as has been done by others.¹² We have independently verified that tissue autofluorescence at 630 nm excitation does not contribute significantly to the spectrally integrated signal by measuring tissue fluorescence prior to the application of the photosensitizer. In fact, the autofluorescence was below the detection threshold of our instrument at 630 nm excitation, given the sensitivity and exposure times employed in our study.

3.1.

Liquid Phantom Study

A series of 16 liquid phantoms were prepared, varying concentrations of India ink (ProArt) as absorber, Intralipid^® (20%, Fresenius Kabi) as scatterer, and PpIX (Spectrum Labs) as fluorescing agent. At each of the four concentrations of fluorophore (1, 3, 5, and 10 μg/ml), four combinations of India ink and intralipid concentrations were employed to provide a range of background optical properties in the presence of the PpIX. Table 1 shows the measured properties from differing concentrations of India ink and Intralipid for each concentration of PpIX. These phantoms were prepared from recipes developed by Kepshire ¹³ Each phantom was poured into a 10 × 8 mm deep cylindrical well surrounded by a diffuse silicone phantom and imaged in both SFDI and fluorescence modes. Since the absorption of India ink is relatively flat over the emission band of PpIX, 〈μ_a 〉_em is assumed to be equivalent to the μ _a calculated at 730 nm; 〈μ_s ^′〉_em, however, was calculated in the same manner as described above.

Table 1

Optical properties of phantoms for fluorescence correction method validation.

Figure 2a shows the average fluorescence signal detected from the phantom well. The error bars represent the variance of the signal as a result of the different optical properties. As expected, the total fluorescence signal exhibited a strong dependence on the optical properties present within the sample. Upon applying the empirical correction, the optical property dependence of the fluorescence signal is dramatically reduced. Figure 2b shows linear correlation between PpIX concentration and the corrected signal. By fitting a linear model to these results, this method can estimate PpIX concentration to ∼0.2 μg/ml across a relevant range in the context of topical ALA uptake in dermatological applications.¹³

Fig. 2

(a) Raw fluorescence signals detected from liquid phantoms with differing concentrations of both absorber (India ink) and scatterers (Intralipid) present. (b) Corrected fluorescence signals based on measured optical properties at excitation and emission wavelengths.

It is worth noting that this method does begin to deviate from a linear response at high fluorophore concentration. This is a consequence of cross-talk in reflectance signal from the fluorescence generated. This instrument does not block fluorescence from contributing to the reflectance signal detected at the excitation wavelength. Given that the fluorescence signal is typically orders of magnitude weaker than the reflectance signal, for most concentrations of fluorophores, this contribution to the reflectance signal is within the noise of the measurement. At concentrations of 10 μg/ml, however, the fluorescence signal contributes ∼1% to 0.5% of the reflectance signal. In terms of SFDI, this detectable contribution will bias the absorption coefficient toward a smaller value. When the detected fluorescence is corrected by these biased optical properties, the model estimates a lower concentration of fluorophore.

3.2.

In Vivo Measurement

To demonstrate the viability of this method in vivo, one healthy subject was consented under an IRB approved study protocol (UCI HS# 2008-6439). In this study, topical ALA (Levulan Kerastick, DUSA Pharmaceuticals), was applied to a 2-cm diameter region of skin in the volar forearm [Fig. 3a]. The subject was imaged 7 hours after the application of the drug. Figure 3b shows the raw fluorescence image of the affected tissue and Fig. 3d shows that the same image corrected for the native tissue optical properties as well as calibrated for the instrument function provided by the liquid phantom validation study. Whereas the image map from the raw fluorescence is presented in terms of arbitrary units, the corrected image map is presented on a scale of μg/ml. This represents an image of drug concentration and distribution. In this particular case study, the drug concentration was determined to be ∼1.6 μg/ml.

Fig. 3

In vivo example. (a) Image of skin tissue (volar forearm) prior to application of topical ALA, (b) raw fluorescence image, 7 h after application of ALA, (scale bar: arbitrary units). (c) Correction map (1/X _1D) based on measured optical properties and (d) corrected fluorescence image (scale bar units: μg/ml).