1. Introduction

Spontaneous Raman spectroscopy is a label free technique that can provide the fingerprints of molecules for identification [1, 2], which has been used to analyze the chemical properties of a broad spectrum of materials [3]. In particular, Raman spectroscopy has been widely recognized as an effective non-invasive technique with great potential for tissue characterization in biomedicine [4–8]. Compared with other optical spectroscopy technologies, Raman spectroscopy is advantageous in high chemical specificity, non-destructive and label free detection and weak scattering background from water, which makes it particularly suitable for tissue measurements. In traditional Raman spectroscopy using commercial micro-Raman systems, surface or near-surface measurements are often conducted in a highly scattering medium such as tissues [9]. Such measurements are inadequate in the measurement of a sample with a layered structure, in which case separate several Raman measurements each sensitive to a different depth need to be taken quickly. Depth sensitive Raman spectroscopy is particularly useful in the examination of epithelial tissues such as the skin [10] or normal tissue overlaying on a tumor such as in breast biopsies [11, 12]. Several different depth sensitive Raman spectroscopy techniques have been proposed up to date [13–17]. Matousek et al. proposed spatial offset Raman spectroscopy (SORS) [13] to measure the Raman spectrum of the inner content of pills that are coated by a non-active protective layer. Similar to depth sensitive fiber-optic probes proposed for fluorescence and diffuse reflectance spectroscopy [18, 19], the separation of individual source and detector fibers is varied to achieve different sensing depths in SORS. Its efficiency of signal collection [14] was then improved by imaging Raman emission onto an annular fiber assembly with two rings, which was demonstrated to be able to measure the transcutaneous Raman spectra of bones in vivo. Later, Keller et al. [15] demonstrated the application of SORS for depth sensitive measurements from an artificial tissue structure consisting of normal human breast tissues of various thicknesses overlying a breast tumor. While SORS has achieved great success, this technique has one important disadvantage that the divergent excitation light out of an optical fiber [13, 15], or a collimated beam in a standard SORS setup [14], or sometimes a convergent beam for illumination but a ring of fibers spatially apart from the illuminated spot [16] for subsurface detection, induces the diffusion of detected light in the medium therefore it suffers from poor depth resolution.

To overcome this disadvantage, non-contact depth sensitive Raman spectroscopy with focused excitation and focused detection will be necessary. The simplest non-contact depth sensitive Raman measurements with focused excitation and focused detection can be achieved by varying the distance between the objective lens and the sample in a Raman microscope, which has been explored in a defocused Raman system [17] and shown effective in detecting subsurface Raman signals. However, this can cause inconvenience in clinical measurements especially in case of in vivo measurements, because it is difficult to precisely control the lens-sample distance in a reproducible manner. To solve this problem, our group has developed an axicon lens based non-contact probe for depth sensitive optical measurements and demonstrated its use in fluorescence spectroscopy [20]. In this probe, the distance between the objective lens and the sample is always fixed. The sensing depth is varied by changing the distance between two axicon lenses inside the optical system, which can be precisely controlled. Both numerical and experimental results [21] indicated that the depth sensitive setup based on the cone shell configuration using the axicon lens can achieve a significantly greater range of depth sensitivity compared to that based on the cone configuration using an objective lens. Khan et al. [22] borrowed this axicon lens-based setup without modification and showed that this setup also works well in depth sensitive Raman spectroscopy. However, one limitation of this setup is that it can measure the Raman spectrum from only one depth at a time so its speed of data acquisition is limited. Given that each Raman spectrum from tissues can take tens of seconds currently, it would be difficult to measure multiple depth sensitive Raman spectra sequentially and still fulfill critical time constraint in the clinical setting. Later we further developed this axicon lens-based depth sensitive technique by incorporating a customized ring-to-line fiber assembly to achieve snapshot optical measurements from multiple depths simultaneously. The new technique was demonstrated in depth sensitive fluorescence spectroscopy from layered tissue phantoms [23] as a proof-of-concept experiment. However, this system cannot be directly transferred to conduct depth sensitive Raman spectroscopy for two reasons. First, Raman signals are much weaker than fluorescence thus the system needs to be redesigned to maximize Raman signals. Second, the depth range needs to be changed to achieve a reasonable depth resolution for typical skin measurements.

In this study, we developed a ray tracing tool to optimize the snapshot depth sensitive setup to achieve the tradeoff between signal collection efficiency and depth resolution for Raman measurements in the skin and the system was built according to the result of optimization. Moreover, the snapshot system was demonstrated to be able to acquire depth sensitive Raman spectra from not only transparent and turbid skin phantoms but also from ex vivo pork tissues and in vivo human thumbnail when the excitation laser power was limited to the maximum permissible exposure (MPE) for human skin. Our results suggest the great potential of depth sensitive Raman spectroscopy in the characterization of the skin and other epithelial tissues in the clinical setting and other similar applications such as quality monitoring of tablets and capsules in pharmaceutical industry requiring the rapid measurement of depth dependent Raman spectra.

2. Methods

2.1 System setup

The optical setup for the proposed snapshot depth sensitive Raman measurements is illustrated in Fig. 1. A near-infrared single-mode laser (XTRA II, Toptica Photonics, Victor, NY, USA) with 785 nm central wavelength is used as the excitation light source. The excitation light is collimated by an FC/APC triplet collimator (TC25APC-780, Thorlabs, Newton, MA, USA), and then the beam diameter (around 7 mm) is adjusted to about 3 mm by a pinhole (SM1D12, Thorlabs, Newton, MA, USA). A laser line filter (LL01-785, Semrock, Rochester, NY, USA) is used to reject the undesired side band. A dichroic mirror (LPD02-785RU, Semrock, Rochester, NY, USA) is used to separate the excitation and emission beams. The excitation beam is focused into the sample and forms a line after passing through an axicon lens (PLANO-CONVEX AXICON, Altechna, Vilnius, Lithuania), in which the length of the focal line depends on the incident beam size and the apex angle of the axicon. Raman light emitted from the focal line in the tissue sample passes the dichroic mirror and then a long pass filter (BLP01-785R-25, Semrock, Rochester, NY, USA) to suppress the excitation light. With the assistance of an objective lens (CFI PLAN ACHROMAT10x/0.25NA, Nikon, Tokyo, Japan), Raman signals emitted from different depths are simultaneously acquired by the proximal end of a fiber assembly, in which each depth is translated to a different radial distance from the center of the fiber assembly as elaborated below. The distal end of the fiber assembly is fed to the input slit of a spectrograph (ACTON LS-785, Princeton Instruments, Trenton, NJ, USA), which is connected to a CCD camera (PIXIS 400BR_eXcelon, Princeton Instrument, Trenton, NJ, USA) for spectral acquisition. The spectral resolution of the spectrograph is 5 cm⁻¹.

 

Fig. 1 Schematic of the snapshot depth sensitive Raman measurements system. AX: axicon; DE: distal end of the fiber assembly; DM: dichroic mirror; LLF: laser line filter; LPF: long pass filter; L₁: collimation lens; L₂: objective lens; PE: Proximal end of the fiber assembly. Note that ray focusing as illustrated in this figure works precisely only up to a depth equivalent to a few transport mean free path lengths.

Download Full Size | PDF

The length of the focal line, denoted as L, is given by

Table 1. Distances of the fiber rings from the center of the dead fiber in the fiber assembly.

View Table

2.2 Numerical simulation and optimization of axicon to fiber ring coupling

In order to simulate and optimize the coupling between the axicon and the fiber ring of the fiber assembly, a ray tracing code was written in Matlab to find the image of those representative points along the focal line on the proximal end of the fiber end for high coupling efficiency as illustrated in Fig. 2(a). It should be pointed out that only half of the light rays are shown for clarity. In Fig. 2(b), which is the inset of the two representative points along the focal line, i.e. $z_{\text{1}}$ and $z_{\text{2}}$, correspond to probing depths of 0.3 mm and 2.1418 mm and map to the first and fifth fiber rings, respectively.

 

Fig. 2 Ray tracing of two points along the focal line imaged by an axicon and a convex lens. Note that only light rays that can reach the convex lens are traced for clarity. Light rays on the left half of the hemisphere out of the two points are omitted for clarity. Light rays propagate from the top to the bottom.

Download Full Size | PDF

Assume that $z$ represents the axial coordinate, and r indicates the lateral coordinate from the center of the axicon as shown in Fig. 2. Note that the edge thickness and the apex angle of the axicon are 5 mm and 110 degrees, respectively, in Fig. 1. A convex lens with a focal length of 20 mm is used to form the images of the two points, which is treated as a thin lens in ray tracing for simplicity without losing generality. Let $z_{\text{1}}\text{'}$ and $z_2\text{'}$ indicate the images of $z_{\text{1}}$ and $z_2$ as shown in Fig. 2(c). It is clear that the two points separated in the $z$ dimension are imaged to two different radial distances from the central axis of the axicon. This finding implies that any single point on the focal line in the $z$ dimension will be transformed to a ring in the r dimension. To achieve high coupling efficiency, the radial distances of the innermost and outermost fiber rings as shown in Table 1 need to match those of the two images, i.e. $z_{\text{1}}\text{'}$ and $z_2\text{'}$, respectively. It was found that the best match can be achieved when the imaging lens is placed at $z=-65$ mm. In short, Fig. 2 confirms that the depth dimension in a sample illuminated by the focal line can be transformed to the radial dimension by an axicon lens to achieve snapshot depth sensitive measurements and the parameters for optimal coupling can be found using the ray tracing code.

From Fig. 2, it can be seen that the two images $z_{\text{1}}\text{'}$ and $z_2\text{'}$ suffer from comatic aberration. This observation suggests that sharp imaging cannot be achieved after an axicon unless the geometric aberration due to light refraction induced by the axicon can be corrected. $z_{\text{1}}\text{'}$ and $z_2\text{'}$ are spaced apart in $z$ dimension by about 1 mm, which requires a relatively large depth of field, i.e. DOF, in the imaging system. As a result, it is difficult to image all rings sharply on the same detector plane. The DOF of the imaging system can be improved by using an imaging lens with a long focal length and a small aperture at the expense of light throughput. Therefore, there is a tradeoff between the depth of field and the light throughput. For Raman measurements, high light throughput is preferable because the Raman signal is weak. A series of simulations were run to determine the axial and lateral locations of the fiber rings on the proximal end of the fiber assembly. The optimal configuration was found when the following conditions were met. First, the radius of the outer ring matches the desired maximal depth after transformation by the axicon lens. Second, the focal length of the imaging lens is set as long as commercially available. Third, the number of light rays out of the two points $z_{\text{1}}$ and $z_2$ that can reach the fiber rings needs to be maximized.

2.3 Depth calibration of the measurement system

Depth calibration was conducted to quantify the actual probing depth of each fiber ring in the final system. For this purpose, fluorescence measurements were taken because of its stronger signal compared to Raman measurements. A layer of Rhodamine 6G (R4127, Sigma-Aldrich, St. Louis, MO, USA) powder with a thickness of 70 μm was used as the test sample, in which the thickness was achieved by clamping the powder between two cover slips. The incident power density reaching the sample was set as 40 mW/cm². The experiment was performed by vertically moving the test sample from the axicon tip to the maximum probing depth with a step size of 3 μm. In all measurements, the exposure time was fixed at 0.5 second and the spectrum was accumulated for 12 times. The depth calibration results are plotted in Fig. 3, in which only the data at a wavelength of 850 nm are displayed and a depth of 0 mm corresponds to 0.3 mm below the axicon tip.

 

Fig. 3 Depth calibration results of fiber rings for signal collection. (a) Measured fluorescence intensity (at 850 nm) of each fiber ring as a function of sample depth; (b) Normalized fluorescence intensity (at 850 nm) of each fiber ring as a function of sample depth.

Download Full Size | PDF

Figure 3(a) shows the measured fluorescence intensity of each fiber ring as a function of sample depth. It can be seen that the collected fluorescence intensity decreases as the fiber ring shifts from the innermost one to the outermost one. The effective probing depth of each fiber ring can be determined easily after the measured fluorescence intensity profile is normalized as shown in Fig. 3(b). The full widths at half maximum (FWHM) from ring 1 to ring 5 are 0.96 mm, 0.96 mm, 0.96 mm, 0.96 mm and 1.02 mm, respectively, according to Fig. 3(b). Moreover, the effective probing depths are approximately 0.06 mm,0.30 mm,0.54 mm, 0.87 mm and 1.20 mm. The depth calibration result in Fig. 3 provides precise depth information in an optically transparent sample. The actual probing depth in a turbid sample would be smaller with poorer depth resolution.

3. Measurements of two-layered transparent and turbid skin phantoms

Ex vivo Raman measurements were carried out on two-layered transparent and turbid skin phantoms to demonstrate the effectiveness of the proposed system. While the former set of phantoms represented the ideal situation, the latter set of phantoms indicated the performance that can be practically achieved.

The two-layered transparent phantoms were made by dissolving urea (V3171, Promega, Madison, WI, USA) and potassium formate (294454-500G, Sigma-Aldrich, St. Louis, MO, USA) in 1.5% (weight/volume) agarose (PC0701-500G, Vivantis, Subang Jaya, Malaysia) to create two separate layers with 1 M urea and 1 M potassium formate, respectively. Then the layer of urea was put on top of the layer of potassium formate to create a two-layered phantom. The thickness of the potassium formate layer was fixed at 3 mm, while the thickness of the urea layer was 0.5 mm or 1 mm for different measurements. During measurements, each phantom was put under the axicon tip with a small air gap around 0.3 mm in between. Due to the fact that urea and potassium formate have two significant Raman peaks at 1000 cm⁻¹ and 1345 cm⁻¹, respectively, the depth sensitivity of Raman measurements performed in each fiber ring can be quantified by calculating the ratio of the two Raman peak intensity values when the thickness of the top layer was changed.

The power density incident on transparent phantom samples was 1.0750 W/cm². The exposure time was set as 5 seconds, and each spectrum was accumulated for 24 times. The Raman spectra measured from all fiber rings are presented in Fig. 4(a) when the top layer thickness of phantoms was 0.5 mm. Fluorescence background has been removed by fitting the raw spectrum to a 5-th order polynomial and then subtracting the fit from the raw spectrum. It can be seen that the intensity of potassium formate peak is smaller than that of urea peak in rings 1 through 3 and the trend reverses in rings 4 and 5. This monotonic trend becomes more obvious in the plot of the peak intensity ratio in Fig. 4(b), in which the intensity ratio of potassium formate peak to urea peak increases from rings 1 to 5. This suggests that the sensitivity of Raman measurements to the bottom layer goes up with the ring number, whichcorresponds to the increasing radial distance of the fiber ring to the center of the fiber assembly. The same trend can be observed when the top layer thickness was increased to 1 mm as shown in Fig. 4(b) except for a smaller range in the peak intensity ratio.

 

Fig. 4 Raman measurements from two-layered transparent phantoms where the top layer contained urea and the bottom layer with a constant thickness of 3 mm contained potassium formate. (a) Raman spectra measured from all fiber rings when the top layer thickness of the phantom was 0.5 mm. (b) Intensity ratio of potassium formate peak to urea peak when the top layer thicknesses of phantoms were 0.5 and 1 mm, respectively.

Download Full Size | PDF

The same experiments were also carried out on turbid phantoms. The turbid phantoms were made by adding polystyrene microspheres (1 μm, Catalog No. 07310, Polysciences, Warrington, PA, USA) into the phantoms to create a polystyrene microsphere concentration of 1.21% by volume. The reduced scattering coefficient ${\mu }_s\text{'}$ of the turbid phantoms was about 3.1 mm⁻¹ at 785 nm, to match the average reduced scattering coefficient of human epidermis and dermis [24]. Similar to Fig. 4(a), it can be seen in Fig. 5(a) that the intensity of potassium formate peak is smaller than that of urea peak in rings 1 through 4 and the trend reverses in ring 5. This monotonic trend becomes more obvious in the plot of the peak intensity ratio in Fig. 5(b), in which the intensity ratio of potassium formate peak to urea peak increases from rings 1 to 5. This suggests that the sensitivity of Raman measurements to the bottom layer goes up with the ring number, which corresponds to the increasing radial distance of the fiber ring to the center of the fiber assembly. The same trend can be observed when the top layer thickness was increased to 1 mm too as shown in Fig. 5(b) except for a smaller range in the peak intensity ratio. Therefore the trend in depth sensitivity for turbid phantoms is in general similar to that for transparent phantoms except that the range of peak intensity ratio is significantly smaller.

 

Fig. 5 Raman measurements from two-layered turbid phantoms where the top layer contained urea and the bottom layer with a constant thickness of 3 mm contained potassium formate. (a) Raman spectra measured from all fiber rings when the top layer thickness of the phantom was 0.5 mm. (b) Intensity ratio of potassium formate peak to urea peak when the top layer thicknesses of phantoms were 0.5 and 1 mm, respectively.

Download Full Size | PDF

Phantom experiments demonstrate that the developed system is effective in the detection of Raman signals from different depths in both optically transparent and turbid samples. In the following sections, we will further demonstrate the effectiveness of the system in the measurements of biological samples including pork tissues and human thumbnail.

4. Measurements of pork tissues ex vivo and human thumbnail in vivo

An untreated pork sample was extracted from a pig (produced from Pulau Bulan, Indonesia). According to American national standard for safe use of lasers (ANSI Z136.1), the maximum permissible exposure (MPE) for the skin is 0.296 W/cm² when the exposure time is varied from 10 seconds to 8 hours for laser light at 785 nm. To verify the Raman spectra to be measured by the proposed system, a commercial micro-Raman system (innoRam-785S, B & W Tek, Newark, DE, USA) was used to measure every individual layer in the same sample by taking measurements from the side of the sample, which served as a reference to validate the Raman measurements from the top surface made by the proposed system. The spectral resolution of the commercial system was around 4 cm⁻¹. The spot size of the laser beam incident on the pork sample was around 0.1 mm, and the beam power density was set to 1 × MPE. The exposure time of the CCD camera was set at 5 seconds and each spectrum was accumulated for 36 times. Therefore, the total exposure time of each pork sample was 3 minutes.

The measured Raman spectral is presented in Fig. 6. The skin in the pork sample is slightly thinner than 1 mm as shown in Fig. 6(a). It’s mostly fat below the skin. The average Raman spectra in a range of 980 cm⁻¹ to 1700 cm⁻¹ measured from the pork sample using the commercial micro-Raman system are presented in Fig. 6(b). Similarly, the fluorescence background of every spectrum has been removed by fitting the raw spectrum to a 5-th order polynomial and subtracted off. Then the resulting spectrum was smoothed by applying low pass Gaussian filtering. In general, biomolecules in tissues contain many molecular bonds that can emit specific Raman peaks upon laser excitation [25–31]. For example, in Fig. 6(b), the strongest Raman peak at 1450 cm⁻¹ can be attributed to CH₂ scissoring mode and the Raman peak at 1654 cm⁻¹ can be attributed to amide I band [28, 30, 31]. It can be also seen that the Raman peak intensity values of the fat layer are in general much stronger than those of the skin layer. Specifically, the fat layer has one significant Raman peak at 1299 cm⁻¹ (attributed to CH₂ twist); while the skin layer has one Raman peak at 1253 cm⁻¹ (attributed to amide III band). Moreover, the Raman peak intensity value of the fat layer at 1299 cm⁻¹ is much higher than that of the skin layer at 1253 cm⁻¹, which indicates the protein content in the skin layer is richer than that in the fat layer [15]. The above information can be utilized to indicate the performance of depth-sensitive measurements in the pork sample.

 

Fig. 6 Raman measurements from a pork sample with 1 × MPE. (a) White light photo of an untreated pork sample measured with a ruler on the side. The space between every two adjacent white ticks is 1 mm. (b) Raman spectra of individual layers measured by a commercial micro-Raman system from the side of the pork sample with 3 minutes exposure time. (c) Depth sensitive Raman spectra of the same pork sample measured by the proposed snapshot system with 2 minutes exposure time.

Download Full Size | PDF

Then the proposed system was employed to measure the same pork sample from the top surface with a power density equal to 1 × MPE. The total exposure time was reduced to 2 minutes. The measured Raman spectra are illustrated in Fig. 6(c), in which fluorescence background has been removed and spectra were smoothed in the same manner as previously.

An obvious observation is that the ratio of the Raman peak intensity at 1299 cm⁻¹ (fat) to that at 1253 cm⁻¹ (skin) increases nearly monotically with the ring number. This implies that the Raman signal detected by a fiber ring with a larger index number is more sensitive to the underlying fat layer and less sensitive to the superficial skin layer. The ratio of two Raman peak intensity values rises drastically to a large value starting from ring 3, which corresponds to a target depth of 0.6 mm. This observation qualitatively agrees well with the skin layer thickness of the pork sample shown in Fig. 6(a).

A similar set of experiments were performed to measure Raman spectra from an adult Asian male’s thumbnail in vivo. Since the maximum probing depth of our snapshot system is about 1.8 mm, the Raman spectra of the tissue immediately below the nail plate will also be detected by the outer fiber rings. It is expected that the depth sensitive Raman spectra of the nail plate and the tissue below it are different and both can be detected by our system. In order to verify such depth sensitive Raman spectra, the Raman spectra from the superficial region in the thumbnail and the skin right next to it were first measured separately by the commercial micro-Raman system. Since the thickness of the epidermis next to the thumbnail was around 0.4 mm (measured by optical coherence tomography), the focal point of the micro-Raman system was adjusted to 1 mm below the surface to reach the dermis when measuring the skin next to it. The power density of the laser beam incident on the thumbnail and the nearby skin was set as 2 × MPE and the total exposure time of each measurement was 3 minutes in these measurements, in order to obtain reference Raman spectra with high signal to noise ratio. In contrast, the power density of the laser beam incident on the top surface of the thumbnail in depth sensitive measurements using the proposed system was set as 1 × MPE and the total exposure time of each measurement was 1.5 minutes.

The measurements results are presented in Fig. 7. The average Raman spectra of the nail plate in the thumbnail and the dermis of the skin right next to it measured by the commercial micro-Raman system are shown in Fig. 7(a). The spectral features match well with the literature [25, 31]. Nearly all major Raman peaks reported in the literature can be identified in Fig. 7(a), such as those at 1003 cm⁻¹ (attributed to phenylalanine), 1125 cm⁻¹ (attributed to C-C/C-N stretching), 1543 cm⁻¹ (attributed to C = C stretching), and 1652 cm⁻¹ (attributed to amide I band). For the nail plate, the Raman peak value at 1344 cm⁻¹ (attributed to CH₂ deformation mode, such as CH₂ twist) is smaller than that at 1453 cm⁻¹ (attributed to CH₂ scissoring mode), while the opposite occurs for the dermis of the skin right next to the thumbnail. This significant difference in spectral feature between the nail plate and the dermis of the nearby skin is also observed in depth sensitive Raman spectra detected from the top surface of the thumbnail by the proposed system as shown in Fig. 7(b). The measured Raman peak value at 1344 cm⁻¹ is smaller than that at 1453 cm⁻¹ for rings 1 through 3, which agrees with the Raman spectrum of the nail plate in Fig. 7(a). In contrast, the measured Raman peak value at 1344 cm⁻¹ is larger than that at 1453 cm⁻¹ for rings 4 and 5, which agrees with the Raman spectrum of the dermis in Fig. 7(a). This observation suggests that the Raman signal detected by rings 1 through 3 are more sensitive to the nail plate, while those detected by rings 4 and 5 are more sensitive to the dermis immediately below the nail plate. The above thumbnail experiment was repeated five times to obtain multiple measurements results to demonstrate the stability of our setup for in vivo applications. The ratio of the peak value at 1344 cm⁻¹ to that at 1453 cm⁻¹ for every ring was plotted with error bars, as shown in Fig. 7(c). It can be easily seen that the ratio increased significantly with the ring number especially after ring 3, which demonstrates the good depth sensitivity of our setup even in consideration of the high local variance of human skin. Given that the effective probing depths of rings 3 and 4 are 0.54 mm and 0.87 mm, respectively, according to Fig. 3, the thickness of the nail plate should be somewhere between 0.54 and 0.87 mm. To verify the thickness of the nail plate, an optical coherence tomography (OCT) system [32, 33] was used to acquire the cross-sectional image of the nail plate and the dermis region below it as shown in Fig. 7(d). The lateral scanning length of OCT was 6.5 mm. It can be seen from the OCT image that the thumbnail thickness was about 0.64 mm (assuming that the refractive index of the nail is 1.52 [34]), which agreed reasonably with the value estimated from depth sensitive Raman spectra.

 

Fig. 7 Raman measurements from an adult Asian male’s thumbnail and the skin next to it in vivo, and the cross-sectional image of the thumbnail region. (a) Raman spectra measured by a commercial micro-Raman system from the nail plate in the thumbnail and the dermis of the skin right next to it, respectively, with 3 minutes exposure time and 2 × MPE. (b) Depth sensitive Raman spectra detected from the top surface of the thumbnail by the proposed snapshot system with 1.5 minutes exposure time and 1 × MPE. (c) Ratio of the Raman peak intensity at 1344 cm⁻¹ to that at 1453 cm⁻¹ as a function of ring number. The error bar indicates the standard deviation of the ratio in each ring. (d) Cross-sectional image of the measured nail plate and the dermis below it acquired by an optical coherence tomography (OCT) system with a lateral scanning length of 6.5 mm, where the dotted line indicates the interface between the nail plate and the dermis.

Download Full Size | PDF

5. Discussion

We have demonstrated in both the ex vivo and in vivo experiments that our snapshot depth sensitive Raman spectroscopy system can simultaneously acquire decent Raman spectra from multiple depths in the pork tissue (ex vivo) and human thumbnail (in vivo) when the incident power density was 1 × MPE. The snapshot measurement leads to the speed advantage in data acquisition. Furthermore, this was achieved without moving any component, which ensured consistent optical coupling and minimized artificial variation in measured Raman spectra. Both features are highly desirable in future clinical measurements.

The numerical simulation tool for ray tracing that we developed played an important role in system optimization to achieve the tradeoff between signal collection efficiency and depth resolution. Here the depth resolution is defined as the FWHM of fluorescence intensity profile as a function of the depth of a thin sample as shown in Fig. 3. The smaller the FWHM is, the higher the depth sensitivity will be. It needs to be pointed out that another parameter, i.e. the increment in the target depth, could be confused with the depth resolution. The increment in the target depth limits the choices of the depth value corresponding to the vertical center of the measured region in each measurement. It can be found from Fig. 3 that the FWHM, i.e. the depth resolution, reaches 0.96 mm, while the increment in the target depth is around 0.30 mm. This depth resolution is effective for measuring human skin and can be tuned to a greater or smaller value if necessary. Compared to the earlier reported depth sensitive Raman setups [13–15], our setup is expected to yield a considerably better depth resolution, for the reason that the focused beam penetrating into tissues can provide more localized light propagation than the diverging light beam [13, 15] from fibers or the collimated beam [14] from a free-space laser. Moreover, the depth resolution of the snapshot system is comparable to the first-generation depth sensitive Raman setup involving three axicon lenses [22], where in the latter setup data acquisition is much slower.

The capability of the axicon lens to transform information on the depth dimension to that on the lateral dimension is the key to the success of the snapshot depth sensitive optical spectroscopy shown here. However, it can be seen from Fig. 2(c) that such transformation is incomplete. The images of two points at two different depths are separated not only in the lateral dimension but also in the depth dimension. While the separation in the lateral dimension is desired, the separation in the depth dimension is not because this requires a relatively large depth of field in a two-dimensional detector. This is one factor limiting the depth resolution of this axicon lens based technique. Another limiting factor is geometric aberration induced by the axicon lens, which is evident from the fact that the two images of $z_{\text{1}}$ and $z_2$, $z_{\text{1}}\text{'}$ and $z_2\text{'}$, are comas instead of infinitesimal points as shown in Fig. 2(c). These two factors make it difficult to achieve a depth resolution better than a few hundred micrometers with this axicon lens based system. In order to achieve a much higher depth resolution, a different mechanism other than an axicon lens that can completely transform the depth dimension to the lateral dimension but does not induce geometric aberration will be necessary. This effort is currently ongoing in our lab. The last limiting factor for depth resolution is the ring-to-line fiber assembly. Currently there are only five rings each with a thickness of a couple of hundred micrometers. It is possible to replace it with another fiber assembly with more rings and smaller ring thickness to yield a higher depth resolution.

Depth sensitive Raman spectroscopy is not meant to perform optical sectioning like confocal microscopy. This technique intends to acquire multiple spectra each sensitive to a different depth thus provides more information for tissue characterization. Light scattering can significantly spread a focal spot. In a highly scattering medium such as a tissue, the depth and depth resolution gradually deviate from the theoretical values with an increase in the target depth because of increase in the focal spot size due to light scattering. Therefore, it is often necessary to validate the depth estimated according to depth sensitive Raman spectra using other techniques with the optical sectioning capability such as OCT as done in this study. To minimize the laser spot in the context of strong light scattering and precisely control the depth, other approaches such as wavefront shaping [35, 36] may help. It should be noted that the concept of depth sensitive measurements in this study relies on ray focusing, which works precisely only up to a depth equivalent to a few transport mean free path lengths, in which photons only undergo a few scattering events. When the depth goes greater, photon diffusion will start to dominate and light focus will become blurred. Consequently, depth sensitivity will be gradually decreasing.

Another challenge for depth sensitive Raman spectroscopy is to extract the pure spectrum of each individual layer. There are a few potential approaches to address this problem. One is to model empirically or semi-analytically the contribution of each layer to the signal detected by every fiber ring and then explore an appropriate data processing algorithm to extract the spectrum, which can be the extension of a sequential estimation method [19] intended for two layers. The other is to apply a method based on wavefront shaping [36] or similar techinques to precisely control the focal spot’s location in each individual layer. However, this approach is still quite slow and requires a sophisticated setup to implement.

Snapshot depth sensitive Raman spectroscopy can have broad applications in tissue measurements, not only for diagnosing early epithelial cancer or cancer margin assessment, but also for performing in vivo analysis and skin depth profiling in cosmetics test in order to detect natural moisturizing factor [37] or harmful substance [38]. This technique can also have large impact in pharmaceutical industry to detect the Raman spectrum of the active ingredient underneath the superficial non-active coating just as SORS.

6. Conclusion

We report a snapshot depth sensitive Raman system based on an axicon lens and a ring-to-line fiber assembly to simultaneously acquire Raman signals emitted from five different depths in the non-contact manner without moving any component. A ray tracing tool was developed to optimize the snapshot depth sensitive setup to achieve the tradeoff between signal collection efficiency and depth resolution for Raman measurements in the skin. The snapshot system was demonstrated to be able to acquire depth sensitive Raman spectra from not only transparent and turbid skin phantoms but also from ex vivo pork tissues and in vivo human thumbnail when the excitation laser power was limited to the maximum permissible exposure for human skin. The results suggest the great potential of snapshot depth sensitive Raman spectroscopy in the characterization of layered tissues such as the skin and other epithelial tissues or for cancer margin assessment in the clinical setting. This technique can also have large impact in pharmaceutical industry to detect the Raman spectrum of the active ingredient underneath the superficial non-active coating for quality monitoring of tablets and capsules where the rapid measurement of depth dependent Raman spectra is required.