2.1.

Instrument Setup

The design of the real-time Raman spectrometer was adapted from an instrument setup for lung cancer in vivo measurement and was described previously.^14,16 In brief, excitation from a 785-nm 300-mW stabilized diode laser (model BRM-785, B&W Tek, Newark, Delaware) was coupled into a bifurcated endoscopic Raman probe. Collected emissions were analyzed with an $f/2.2$ spectrograph (model Holospec, Kaiser Optical Systems, Ann Arbor, Michigan) working from 1531 to $3416{\mathrm{cm}}^{-1}$. The power of the excitation light at the sample was around 140 mW. An NIR optimized, back-illuminated, deep depletion charge coupled device (CCD) with $400\times 1340\text{pixels}$ ($20\times 20\mu \mathrm{m}\text{per}\text{pixel}$) was used for detection (model Spec-10:400BR/LN, Princeton Instruments, Trenton, New Jersey). The 3-m-long Raman probe (2.3 mm in diameter) was enclosed in a stainless steel monocoil to give it increased tensile strength while maintaining a high degree of flexibility. To protect the probe from biological contaminants, it was covered with a medical-grade heat shrink tubing (Iridium, Cobalt Polymers, Cloverdale, California). The central $200\text{-}\mu \mathrm{m}$-diameter lone delivery fiber was surrounded by 34 collection fibers ($100\mu \mathrm{m}$ in diameter). Another specially designed optic fiber bundle was employed to couple collected light to the spectrograph.¹⁷ Wavelength calibration was done with mercury argon (Hg/Ar) and xenon (Xe) calibration lamps (Ocean Optics, Dunedin, Florida) and spectral resolution was estimated to be $\approx 6{\mathrm{cm}}^{-1}$ at 1013 nm (corresponding to a Raman shift of $2867{\mathrm{cm}}^{-1}$ with 785-nm excitation).

2.2.

Sheath Design

A low-cost sheath material needs to be biocompatible, mechanically able to meet the required needs, and contribute minimal interfering fluorescence or Raman emissions. The material chosen was polyvinylidene difluoride (PVDF), a transparent, slightly flexible material with high tensile strength that can be manufactured into tubes of different sizes. PVDF is biocompatible and well suited for biomedical applications. It does have some Raman features, but since the material is not directly placed in the excitation and collection optical paths (not used as the window), it was not expected to have a significant effect on the tissue material Raman. Specifically in the fingerprint region, and below $1000{\mathrm{cm}}^{-1}$, PVDF has Raman emission mainly dominated by two peaks¹⁸ at 795 and $839{\mathrm{cm}}^{-1}$. Between 1000 and $1800{\mathrm{cm}}^{-1}$, the PVDF generates an intense Raman band at $\sim 1433{\mathrm{cm}}^{-1}$, originating from the ${\mathrm{CH}}_2$ bending mode, atop of a flat featureless spectral background. In the high wavenumber region ($\sim 2400$ to $3800{\mathrm{cm}}^{-1}$), a relatively strong Raman peak around $3000{\mathrm{cm}}^{-1}$ exists.¹⁹

A PVDF tube with a 0.15-mm wall thickness has a good balance between flexibility and rigidity as well as an intermediate value in terms of coefficient of friction when compared to other available materials. To maintain a fixed distance between the Raman probe tip and the tissue surface, an optical spacer (window) was designed and attached to one end of the PVDF tube using medical-grade epoxy (Epotek 301, Epoxy Technology Inc., Billerica, Massachusetts). The length of the spacer was designed according to the illumination spot size originally calculated in previous research, and taking into account the refractive index of the material. Both entrance and exiting surfaces of the optical spacer had an antireflective coating on them. The disposable sheath and Raman probe assembly is shown in Fig. 1.

Fig. 1

The Raman probe and disposable sheath assembly with a Canadian dime (diameter: 18.03 mm) for scale. Inset: zoomed-in view of the optical spacer at the front end.

2.3.

Data Handling and Software

A commercial software product, WinSpec/32 (Version 2.5.20.2, Princeton Instruments), was used to operate the spectrometer and acquire Raman spectra. The CCD pixel rows were binned orthogonally to the wavenumber axis to enhance the SNR of the spectra.¹⁷ A 13-point third-order Savitzky-Golay smoothing filter using built-in functions in LabVIEW (National Instruments, Austin, Texas) was applied to smooth the Raman spectra.²⁰ The CCD dark noise background was subtracted from acquired spectra.²¹ Fluorescence background was also removed using a fifth-order polynomial fitting algorithm with a window size of 5 and a stopping criteria of $<5\%$ between the standard deviation of fitting residuals for two iterations.²²

3.1.

Spectral Property of Spacer Glass

Because both excitation and emission light transmit directly through the optical window at the sheath tip, the window material must contribute only a minimal amount of spectral interference to the Raman measurement. In addition, the selected optical material needs to be biocompatible and economically affordable. Quartz and fused silica windows meet the required biocompatibility and cost requirements. To test spectral interference levels incurred by these materials under laser radiation, in vivo Raman spectra of palm skin from a volunteer were collected with either optical-grade quartz or fused silica windows (1 cm in thickness) placed in between the Raman optic fiber probe and the skin surface (Fig. 2). The spectra were first normalized to a common peak at $1657{\mathrm{cm}}^{-1}$ and then shifted vertically for clarity. As is evident in Fig. 2, the Raman spectrum measured through quartz window was accompanied by strong fluorescence between 1690 and $2500{\mathrm{cm}}^{-1}$, which overwhelmed the Raman peak at $1657{\mathrm{cm}}^{-1}$. The spectral difference in Fig. 2(b) shows the strong fluorescence (1690 and $2500{\mathrm{cm}}^{-1}$) for the quartz window and its reduced transmission that weakens the skin Raman peak intensity between 2800 and $3000{\mathrm{cm}}^{-1}$. The difference between normalized Raman spectra acquired with the probe alone and with the fused silica window is close to zero across the entire spectral region [Fig. 2(b)]. Therefore, the fused silica window in the optical path did not interfere with Raman measurement significantly and Raman peaks pertaining to skin tissues could be preserved with a minimum amount of distortion. Therefore, fused silica is considered the preferred material for constructing the probe sheath tip in this application.

Fig. 2

In vivo Raman skin spectra collected with quartz, fused silica, or air between palm skin surface and optic fiber probe: (a) raw Raman spectra and (b) difference spectra between normalized Raman spectra acquired with and without optical window.

3.2.

Spectral Property of Epoxy and Polyvinylidene Difluoride Sheath Wall

When the Raman probe is sheathed, excitation laser light reflected from either the fused silica window surfaces or the tissue surface could return to the sheath and generate unwanted background signals. Although this interference is expected to be minimal, it was important to confirm this by measuring the Raman spectra of the epoxy and the sheath material. To test the epoxy, a 1-mm-thick epoxy resin was deposited onto a piece of anodized aluminum foil and allowed to dry for 48 h. To test the sheath, multiple small sections of the wall were cut from a piece of PVDF tubing and stacked to a thickness of 0.6 mm on top of a piece of anodized aluminum foil. The first experiment on these samples used the same measurement conditions as for in vivo human tissue with the following measurement parameters: excitation power 140 mW, integration time 1 s, and probe to sample distance 7.5 mm. The second experiment on these samples was designed to clearly identify unique Raman features for epoxy and PVDF by increasing the amount of collected light. To achieve this, the sampling distance was reduced to 0.5 mm and integration time was set to 10 s with the same excitation power as in the in vivo measurements. The spectra of anodized aluminum foil was also acquired as the background and subtracted from the Raman spectra for PVDF and epoxy. Figure 3(a) shows the raw Raman spectra for skin from a volunteer, the epoxy, and the PVDF tubing stack obtained in the first experiment. Apparently, skin generated the highest level of fluorescence background compared to sheath materials. The sharp peak at $1657{\mathrm{cm}}^{-1}$ and Raman band between 2800 and $2940{\mathrm{cm}}^{-1}$, which belong to skin, could be easily identified. There was a small peak around $3000{\mathrm{cm}}^{-1}$ for PVDF and a peak at $1613{\mathrm{cm}}^{-1}$ for epoxy resin. The common peak around $1780{\mathrm{cm}}^{-1}$ in epoxy and PVDF spectra was believed to belong to the optic fiber. Figure 3(b) shows the results obtained in the second experiment. In this case, the Raman peaks for epoxy and PVDF are greatly enhanced with optimized sampling geometry and 10 times integration time. The epoxy showed a sharp spectra peak at $1613{\mathrm{cm}}^{-1}$ and a relatively broad peak from 2850 to $2950{\mathrm{cm}}^{-1}$. The latter overlaps with the Raman features of skin, as does the Raman peak around $3000{\mathrm{cm}}^{-1}$ from the PVDF, which is in agreement with previous reports.¹⁹ Although easily distinguishable, it should be noted that the Raman spectra for both PVDF and epoxy as shown in Fig. 3(a) were obtained with full excitation power and with a geometry that maximized the collection of any emission from these samples. Under actual tissue measurement conditions [Fig. 3(a), top curve], both PVDF and epoxy were only exposed to laser light specularly reflected from tissue and spacer window surfaces, which accounted for only 1 to 2% of the total excitation power, and any excitation of these materials would occur outside of the aperture for direct collection by the Raman probe fibers. Furthermore, both the epoxy and PVDF were also at least eight times thinner in the assembled sheath, which would also lead to a much lower background signal from these materials. Thus, despite their clear Raman features [Fig. 3(a), bottom and middle curves], the spectral contribution from PVDF and epoxy to live tissue measurement could be considered minimal.

Fig. 3

Raman spectra of PVDF (sheath wall) and epoxy resin deposited on aluminum foil: (a) raw Raman spectra for palm skin, epoxy resin, and PVDF polymer as measured in experiment 1 and (b) Raman spectra of epoxy resin and PVDF on top of anodized aluminum foil in experiment 2. The spectrum of anodized aluminum foil was subtracted from Raman spectra of PVDF and epoxy. Spectra were shifted on the intensity scale for clarity.

3.3.

Signal Collection Efficiency

It is of interest to find out how the signal collection efficiency varies with sampling distance between the tissue surface and Raman probe end. A Tylenol tablet (acetaminophen) was placed underneath the Raman probe and moved vertically by a micrometer sampling stage. The total excitation power from the laser was kept constant. The distance was gradually increased from 0.1 to 10 mm. Three major Raman peaks at 1648, 2931, and $3064{\mathrm{cm}}^{-1}$ were selected [Fig. 4(a)]. The intensity versus sampling distance of the three Raman peaks after fluorescence background removal is shown in Fig. 4(b).

Fig. 4

Variation in collected spectral intensity of Raman peaks: (a) Raman spectra of acetaminophen and (b) variation in intensity of peaks at 1648, 2931, and $3064{\mathrm{cm}}^{-1}$ with changes in distance between Raman probe and the Tylenol tablet.

As is seen in Fig. 4, the collected Raman peak signal reaches its maximum point when the sampling distance is $\sim 0.5\mathrm{mm}$. Thus, the sampling distance of 0.5 mm suggested that the collection is most efficient for such Raman probe design. However, this does not take into account the multiple scattering events that occur in translucent tissue samples, which would result in a collection efficiency maximum at a greater probe distance. One drawback with short probe to tissue distances is the illumination spot is only $\sim 0.04{\mathrm{mm}}^2$, which is too small to be able to acquire representative Raman spectra of the tumor tissue volume. In addition, the excitation power would have to be reduced drastically to comply with the MPE limit for skin set out in the ANSI guidelines. With an illumination spot size of diameter 3.5 mm, a spectrum is obtained from a tissue volume that is more representative of the bulk tissue and a much higher excitation power can be used to compensate for the greater probe to tissue distance. Therefore, the sampling distance was optimally set at 7.95 mm to produce a 3.5-mm-diameter excitation spot, and an excitation power of 140 mW used here complied with the MPE limit for human epidermis ($1.628\mathrm{J}/\mathrm{cm}$ with 785 nm excitation and an integration time of 1 s). Without the sheath, any slight deviation from the sampling distance toward the tissue site will lead to patients being exposed to a higher dose of laser radiation. Technically, it was challenging to maintain a proper sampling distance during an endoscopic exam where the physician could only make a best judgment about sampling distance based on white light imaging. Therefore, the presence of the Raman sheath ensures that the MPE value of the laser radiation on human epidermis is not exceeded.

3.4.

Spectra Repeatability

The Raman peak intensity tends to vary greatly when the sampling distance changes [Fig. 4(b)]. The peak intensity variation with changing sampling distance unnecessarily complicates endoscopic Raman measurement by introducing measurement errors. Such variation was partially caused by collection efficiency as well as by the change in excited tumor volume. In the endoscope, the only method to adjust the sampling distance was to pull/push the Raman fiber optic bundle through the instrument channel. Therefore, the lack of precision in controlling the sampling distance was anticipated under most operating circumstances. In Fig. 5, the variation in acquired Raman spectra with and without the disposable Raman sheath is demonstrated. Each type of spectra measurement was repeated 20 times by the same operator on the same skin site (human hand palm). The unsheathed Raman probe was held at a distance between 6 and 8 mm away from the skin surface, mimicking the measurement conditions during endoscopic measurement.

Fig. 5

The mean and standard deviation of measured Raman spectra of palm skin using Raman probe with and without the designed disposable sheath. Spectra were intentionally shifted on the intensity scale for clarity.

Figure 5 shows that the spectral variation due to varying sampling distance could be effectively reduced by using the disposable sheath during measurement. On average, the coefficient of variation could be reduced by 1.5 to 8.5 times depending on the spectral range by simply using the sheath.