Optical diagnosis and characterization of dental caries with polarization-resolved hyperspectral stimulated Raman scattering microscopy

: We report the utility of a rapid polarization-resolved hyperspectral stimulated Raman scattering (SRS) imaging technique developed for optical diagnosis and characterization of dental caries in the tooth. Hyperspectral SRS images (512 × 512 pixels) of the tooth covering both the fingerprint (800-1800 cm − 1 ) and high-wavenumber (2800-3600 cm − 1 ) regions can be acquired within 15 minutes, which is at least 10 3 faster in imaging speed than confocal Raman mapping. Hyperspectral SRS imaging uncovers the biochemical distributions and variations across the carious enamel in the tooth. SRS imaging shows that compared to the sound enamel, the mineral content in the body of lesion decreases by 55%; while increasing up to 110% in the surface zone, indicating the formation of a hyper-mineralized layer due to the remineralization process. Further polarized SRS imaging shows that the depolarization ratios of hydroxyapatite crystals ( ν 1 -PO 4 3- of SRS at 959 cm − 1 ) of the tooth in the sound enamel, translucent zone, body of lesion and the surface zone are 0.035 ± 0.01, 0.052 ± 0.02, 0.314 ± 0.1, 0.038 ± 0.02, respectively, providing a new diagnostic criterion for discriminating carious lesions from sound enamel in the teeth. This work demonstrates for the first time that the polarization-resolved hyperspectral SRS imaging technique can be used for quantitatively determining tooth mineralization levels and discriminating carious lesions of body of lesion has higher optical heterogeneity compared with sound enamel and surface zone. The above results demonstrate that the multimodal nonlinear optical imaging technique developed is a powerful tool for providing deep insights into the 3-D morphological architectures and biomolecules/biochemical distributions of the tooth with high spatial resolution. The 3-D imaging ability of the multimodal nonlinear optical imaging technique developed could further facilitate early detection and diagnosis of dental caries beneath the enamel surface.


Introduction
Dental caries is caused by an imbalance between the demineralization and re-mineralization processes of the tooth related to the acid generated from the bacterial activity [1]. Early detection and diagnosis of early dental caries in the tooth is essential to timely intervene the carious process through preventative treatments [2]. The visual and tactile inspection method that is commonly used to detect dental caries lacks sufficient sensitivity for detecting early onset of caries. Moreover, the use of sharp dental explorer during the visual and tactile examination may result in irreversible traumatic defects on incipient lesion [3]. Other diagnostic techniques, such as fiber optic trans-illumination [4], quantitative light-induced fluorescence [5], have problem with high false positive or false negative rates for detecting early stage of dental caries. With a high biochemical/biomolecular specificity and biomolecular orientation sensitivity, polarized Raman spectroscopy and microscopy has emerged as an appealing tool for biochemical characterization and diagnosis in the tooth [6][7][8]. However, due to the extremely weak tissue Raman scattering process, the measurements of tissue micro-Raman images may take tens of hours, hampering its wide applications in rapid diagnosis and characterization of biomedical tissues. The concomitant strong fluorescence background from tooth samples may also overwhelm weak tissue Raman signals [9], making micro-Raman imaging on dental caries even more challenging. To address the above problems encountered in confocal Raman imaging, coherent Raman scattering (CRS) (e.g., coherent anti-Stokes Raman scattering (CARS) and stimulated Raman scattering (SRS)) has been recently developed for biomedical imaging [10][11][12][13][14]. CRS enhances the weak Raman signal by 6 orders of magnitude through resonant enhancement nonlinear processes, enabling CRS imaging speed up to video rate [15,16]. Unlike CARS, SRS is free from nonresonant background interference and has a linear dependence on the biochemical concentration, making SRS to be an attractive tool for quantitative imaging of biochemical compositions and distributions in unstained live cells and tissue [11]. In this work, we report the utility of a rapid polarization-resolved hyperspectral SRS imaging technique developed for optical diagnosis and characterization of carious lesions in the tooth without labeling. We demonstrate that the hyperspectral SRS imaging can acquire Raman images of the tooth covering both the fingerprint (FP) (800-1800 cm −1 ) and high-wavenumber (HW) (2800-3600 cm −1 ) regions within 15 mins, offering 10 3 faster in imaging speed than conventional confocal Raman imaging. The complementary information acquired in both the FP and HW Raman regions can further improve the understanding of biochemical/biomolecular distributions and orientations associated with carious lesions in the tooth.

Tooth samples preparation
This research protocol was approved by the Institutional Review Board (IRB) of National University of Singapore. Without identifiers, the teeth samples donated from various individuals/clinics were washed in distilled water and stored in saline. The teeth were then cleaned using a soft toothbrush and transferred to a 0.1% thymol storage medium before being sectioned longitudinally from cusp tip to cemento enamel junction (CEJ). Sections were prepared in the labial/buccal-lingual/palatal direction and centred through the unworn cuspal tips and the underlying dentine horns, using a Buehler IsoMet 1000 with a cutting diamondwafering blade. With approximately 150-180 μm in thickness, the sections were hand ground using a graded series of Buehler Met-II grinding pads (P800, P1000, P1200, P2500, P4000) with silicon carbide abrasive on a Buehler Phoenix Beta Grinding/Polishing Machine, until a thickness of 80-100 μm was attained and confirmed with vernier calipers. Sections were then washed using distilled water, and air dried for 24 hours to remove smear layer and contaminants from the surface, before preliminary characterization using Olympus BX51 polarized light microscope with a digital microscope camera (Olympus DP25) and imaging software (Olympus Cell D) before polarization-resolved hyperspectral SRS imaging.. Figure 1 shows the schematic of the polarization-resolved hyperspectral SRS imaging system developed for label-free molecular imaging [18]. The excitation laser sources consist of an Nd:YVO4 laser (High-Q Laser, Austria) and an optical parametric oscillator (OPO) (Levante Emerald, APE-Berlin). A portion (20%) of the fundamental output of Nd:YVO4 laser (80 MHz, 7.5 ps pulses at 1064 nm) serves as the Stokes beam for SRS imaging. The remaining output (80%) is frequency doubled (532 nm) to pump the OPO, whereby the tunable signal generation ranging from 670 to 980 nm is used as the pump beam for SRS imaging. The wide tunable range of OPO enables the SRS imaging in both fingerprint (800-1800 cm −1 ) and highwavenumber (2800-3600 cm −1 ) spectral regions. The 1064 nm Stokes beam is modulated at 20 MHz by an electro-optic modulator (EOM). The spatially and temporally overlapped pump and Stokes beams are sent to a multiphoton scanning microscope and focused onto the sample through a water-immersion objective (XLUMPLFLN 20 × , NA = 1.0, Olympus Inc.). The transmitted pump beam is collected in the forward direction with a condenser (NA = 1.4, Nikon Inc.) and spectrally isolated from the Stokes beam with bandpass filter sets (a filter set centered at 780 nm with a 160 nm bandwidth for SRS imaging in HW region; another filter set centered at 900 nm with a 100 nm bandwidth is used for SRS imaging in FP region) for hyperspectral SRS imaging. The modulation of the pump beam intensity due to the stimulated Raman loss (SRL) process is detected by a photodiode (FDS1010, Thorlabs Inc.). A lock-in amplifier with the time constant as short as 100 ns is used to demodulate the pump beam to acquire the SRS signal. SRS images (e.g., 959 cm −1 (ν 1 -PO 4 3of hydroxyapatite (HA) crystals of the tooth) [19]) of 512 × 512 pixels can be acquired within 0.3 s/frame with a 1μs pixel dwell time. The SRS imaging speed can be further boosted to video rate by using resonant scanners. One notes that other nonlinear microscopy imaging modalities (e.g., secondharmonic generation (SHG), third-harmonic generation (THG), two-photon excitation fluorescence (TPEF), coherent anti-Stokes Raman scattering (CARS)) can also be incorporated into the hyperspectral SRS imaging system ( Fig. 1) to better understand the morphological architectures, biochemical structures and compositions of tissue in a comprehensive fashion [20,21]. The choices of photomultiplier tubes (PMTs), filters and dichroic mirrors for epi-detected SHG/THG/TPEF/CARS imaging modalities have been reported in our previous work [22]. For comparison purpose, a confocal Raman microscope (Renishaw inVia, UK) equipped with a 20 × objective (N.A. = 1.0) and a 785 nm laser with an excitation power of 50 mW is utilized for micro-Raman spectroscopy and imaging of the tooth. The spectral resolution of the confocal Raman microscope is ~4 cm −1 .The pixel dwell time is typically >5 s to ensure the good SNRs of Raman spectra for tooth Raman imaging.

Polarization-resolved hyperspectral SRS imaging system
Hyperspectral SRS imaging is accomplished by scanning the pump beam wavelengths of OPO through computer control of the crystal temperature, tilt angles of Lyot filter and the cavity length of OPO, generating a three-dimensional data stack (x, y, Ω), where Ω = ωp-ωs is Raman shift. The tuning of pump wavelengths is synchronized to the frame trigger of microscope to achieve automatically spectral scanning of SRS images at different Raman shifts. The hyperspectral SRS stack is then normalized by the intensity of the pump beam at each wavelength recorded by a photodiode which detects a small fraction of the OPO output after a beam splitter. Automated wavelength tuning, SRS images acquisition, intensity normalization, and three-dimensional hyperspectral SRS data generation are controlled by using a house-built software with LabVIEW programming.
For polarization-resolved SRS imaging, the polarizations of the pump and Stokes beams are rotated independently by using achromatic half-wave plates mounted in step motors. An analyzer is placed before the photodiode with the analyzer polarization parallel to the pump beam polarization. Polarization-resolved SRS images as a function of the polarization angle (θ) between the polarization directions of the pump and Stokes beams are acquired from 0° to 180° with 20° intervals. The total acquisition time of the polarized SRS image (512 × 512 pixels) is typically <30 s.

Image data processing and visualization
Hyperspectral SRS and TPEF/THG images acquired from the tooth were imported into ImageJ (National Institutes of Health) for image processing and video visualization. From the polarized SRS images acquired, the depolarization ratios of specific SRS Raman peaks (e.g., SRS at 959 cm −1 (ν 1 -PO 4 3of HA crystals)) can be calculated by dividing the polarized SRS intensity (I SRS (θ = 90°)) by the polarized SRS intensity (I SRS (θ = 0°)).

Hyperspectral SRS imaging of carious lesions in the tooth
In this study, we measured the hyperspectral SRS images of dental caries covering both the FP (800-1800 cm −1 ) and HW (2800-3600 cm −1 ) regions with a spectral interval of 10 cm −1 . The Raman spectra of tooth measured with confocal Raman microscope show that the Raman peaks of tooth typically have spectral widths of ~15 to 20 cm −1 . Hence, a scanning spectral interval of 10 cm −1 is used to acquire hyperspectral SRS images of the tooth. The entire hyperspectral SRS image stacks covering the 800-3600 cm −1 range (512 × 512 pixels × 200 Raman shifts) can be acquired within 15 mins (equivalent to Raman microspectroscopy with pixel dwell time of 3 ms/pixel under the excitation powers of 25 mW and 50 mW for the pump and Stokes beams, respectively), offering ~10 3 faster in imaging speed as compared to confocal Raman microspectroscopy (5 s/pixel) under a similar excitation power. Note that if only a single Raman shift is required (e.g., 959 cm −1 (ν 1 -PO 4 3-) reflecting the mineral contents in tooth) for Raman imaging, SRS image can be acquired within 0.3 s (~1 μs pixel dwell time), which is ~10 6 faster in imaging speed than confocal Raman imaging. Figure 2(a) is the polarized light microscope image of dental caries, displaying the three different anatomical zones of enamel caries (i.e., translucent zone, body of lesion and surface zone) in the tooth. Figure 2(b) is the SRS image acquired at 959 cm −1 (ν 1 -PO 4 3of HA crystals), reflecting the distribution of mineral contents in the tooth. SRS intensity decreases from sound enamel to the body of lesion, but shows an increase in the surface zone. The distributions of phosphate across the enamel caries are shown quantitatively in Fig. 2(c) (Intensity profile of SRS at 959 cm −1 along the line indicated in Figs. 2(a), 2(b)). The intensity profile is normalized to its maximum value at surface zone. SRS intensity in the surface zone (Region I) is highest among the four regions, due to the re-mineralization process occurring in the surface zone which results in the hyper-mineralization of the tooth. In the body of lesion, SRS intensity is lowest among the four regions with the value of approximately 0.5, indicating the highest dissolution of mineral components in this zone. SRS signal intensity increases from 0.5 to 0.9 in the translucent zone (region III), indicating the gradual loss of minerals in the translucent zone which is the advancing front of carious process [23]. The changes of the phosphate contents in the dental carious identified by SRS imaging technique in this study are in agreement with histological examination and chemical analysis [24,25]. Figure 2(d) shows the SRS signal at 1070 cm −1 (B type ν 1 -CO 3 2-) related to the carbonate distributions across the tooth. Unlike SRS signal at 959 cm −1 , SRS signal at 1070 cm −1 in the surface zone does not increase to a level higher than the sound enamel due to the loss of carbonate in the demineralization process and the reduced carbonate uptake during the remineralization process [24]. Almost no SRS signal at 2935 cm −1 (CH 3 stretching of proteins) is observed in SRS images (data not shown), reflecting the lack of organic materials existing in enamel caries. Besides SRS imaging, we can also simultaneously measure other nonlinear microscopy images (e.g., SHG, TPEF and THG) on the same tooth samples. For instance, stronger THG signals (reflection of optical heterogeneity) are observed in the body of lesion and translucent zone compared to the sound enamel and surface zone in the tooth [ Fig. 2(e)], indicating the increase of the optical heterogeneity inside the enamel rods in the body of lesion and translucent zone due to the carious process. But a stronger TPEF signal (arising from endogenous fluorophores) [ Fig. 2(f)] is largely observed in the surface zone, indicating a higher content of endogenous fluorophores appearing in surface zone. Therefore, the simultaneous multimodal nonlinear microscopy imaging (SRS/SHG/THG/TPEF) developed in this work further the understanding of biochemical/biomolecule distributions and morphology changes associated with carious transformation in the tooth in a comprehensive way without labeling. SRS spectra from different locations of the tooth can be readily generated from different pixels of hyperspectral SRS images acquired. Figure 3(a) shows the representative SRS spectra of the four regions (i.e., sound enamel, translucent zone, body of lesion, and the surface zone) in enamel caries. Unlike the spontaneous Raman spectrum that is concomitant with strong fluorescence background interference [9], SRS spectrum is free of fluorescence contribution. The prominent Raman peaks at 959 cm −1 (ν 1 -PO 4 3of HA crystals), 1043 cm −1 (ν 3 -PO 4 3of HA crystals), 1070 cm −1 (B type ν 1 -CO 3 2of HA crystals), and 3550 cm −1 (O-H vibration of HA crystals) can be clearly identified in SRS spectra of the tooth. SRS peak intensities at 959 cm −1 (I 959 ) in the body of lesion and translucent zone are much lower than that in the sound enamel, while I 959 in the surface zone is slightly higher compared to the sound enamel. The intensity ratio of SRS at 1043 cm −1 (I 1043 ) to I 959 is highest in the body of lesion (~0.1) and the ratio in the surface zone (~0.03) is slightly higher than that in the sound enamel (~0.025). Raman peak at 1043 cm −1 is stronger than the peak at 1070 cm −1 in the surface zone, translucent zone and body of lesion, while the peak at 1043 cm −1 is less obvious compared with the peak at 1070 cm −1 in the sound enamel. The change of I 1043 relative to I 959 or I 1070 could be due to the enamel crystallite morphology and/or orientation changes caused by the caries [6]. SRS peak at 3550 cm −1 (O-H vibration of HA crystals) is most obvious in the sound enamel, and the peak disappears in the body of lesion and surface zone, indicating the loss of hydroxyl in these two regions due to the breakdown of HA crystals. The SRS spectral differences between the sound enamel and surface zone demonstrate that the remineralization process not only recovers mineral loss in the surface zone, but also results in the biochemical and crystal structural changes that help prevent the tooth from further acid corrosion. TPEF signal enhancement in the surface zone thus could be related to the molecular and morphological changes induced by the remineralization process. Due to the increase of SRS signal intensity at 1043 cm −1 , the distribution of body of lesion can be highlighted by using SRS intensity ratio (I 1043 /I 959 ) mapping [ Fig. 3(b)]. Figure 3(c) is the intensity profile of I 1043 /I 959 along the line indicated in Fig. 3(b), which quantitatively shows the change of the intensity ratio across the enamel caries. I 1043 /I 959 gradually increases from the sound enamel (region IV) to the center of the body of lesion (region II) and reduces towards the surface zone (region I). The above results demonstrate that hyperspectral SRS imaging technique allows rapid molecular mappings of caries and surrounding sound enamel, providing deep insights into the understanding of biochemical changes in dental caries with high spatial resolution and full Raman spectral information.  Fig. 2(a). (c) Intensity profile of I 1043 /I 959 along the line in Fig. 3(b). Regions I-IV correspond to surface zone, body of lesion, translucent zone, and the sound enamel of the tooth, respectively.

Polarization-resolved SRS imaging of enamel lesions
We further investigate the changes of molecular orientations/organizations of the enamel caries by using polarization-resolved SRS imaging technique. Figure 4(a) shows the SRS image at 959 cm −1 of enamel caries whereby different regions of enamel caries can be identified with the pump and stokes beams having the same polarizations. Figure 4(b) shows the representative polarization-resolved SRS intensities at 959 cm −1 in the sound enamel, translucent zone, body of lesion, and surface zone, respectively. Overall, the polarized SRS intensities gradually decrease with the polarization angles (θ) increasing from 0° to 90°, while increase with the angle increasing from 90° to 180°. The SRS intensity in the sound enamel is found to be most sensitive to the polarization angle's change, while least sensitive in the body of lesion. Figure 4(c) shows the mapping of SRS depolarization ratios at 959 cm −1 across the enamel in the tooth. We can easily identify the body of lesion in the SRS depolarization ratio mapping. Clearly, different depolarization ratio values are observed in the dental caries with the body of lesion being the biggest, the translucent zone being in between, while the body of lesion and sound enamel lowest. More specifically, Fig. 4(d) shows the calculated depolarization ratios at 959 cm −1 of the four regions. The mean depolarization ratios of the 5 tooth samples are 0.035 ± 0.01, 0.052 ± 0.02, 0.314 ± 0.1, and 0.038 ± 0.02, respectively, for the sound enamel, translucent zone, body of lesion and the surface zone. The paired two-sided Student's t-test indicates that the difference in depolarization ratios between the sound enamel and the body of lesion are statistically significant with p < 0.00001. The increase of the depolarization ratio indicates the breakage of the vibration symmetry, which could be due to the increased scattering and alteration in the degree of HA crystal orientations caused by the caries. Different regions (translucent zone, body of lesion, surface zone) of enamel caries can be resolved from the polarized SRS image with sub-micron resolution, making polarized SRS imaging suited for determining the extent of carious process. The above results demonstrate that the depolarization ratio mapping obtained from polarized SRS images could be a potential diagnostic criterion for quantitative analysis of dental caries in dentistry.

3-D multimodal nonlinear microscopy imaging of DEJ and enamel lesions in the tooth
The developed SRS imaging technique integrated with other nonlinear optical microscopic imaging modalities (e.g., SHG/THG/TPEF) containing complementary diagnostic information could provide us a deeper understanding of biochemical compositions and morphological structures of tooth in a very comprehensive way. Multimodal nonlinear optical microscopy can be readily applied for 3-D optical sectioning of tooth samples with a high spatial resolution. Figure 5 (Visualization 1) shows the representative 3-D colocalized multimodal nonlinear microscopic images of SRS at 2935 cm −1 (CH 3 stretching of proteins), SHG (collagen), TPEF (endogenous fluorophores), and THG (optical heterogeneity), in the dentinenamel junction (DEJ) vicinity of the tooth with an imaging depth of 72 μm. SRS image at 2935 cm −1 visualizes the distribution of proteins content in the tooth, which could be useful for diagnosis of dental caries. The DEJ visualized by SRS and SHG imaging exhibits a 3-D scalloped appearance with the convexity directed toward the dentin while the concavity toward the enamel. The 3-D structures of twisted dentinal tubules, enamel spindles and enamel rods can be observed in THG imaging. The morphology of the transitional zone between the dentin and enamel that contains inorganic fluorophores varies across different tissue depths as shown in 3-D TPEF images. The results demonstrate that complementary information by each individual imaging modality can provide us with more detailed morphological, biochemical and biomolecular structures and conformation of the tooth.  The morphology of enamel rods and the distribution of enamel caries at different depth can be clearly resolved from both SRS at 959 cm −1 and THG signals. The enamel rods are loosely packaged in the dental caries lesion, which is due to easier dissolution of minerals in the inter-prismatic region [24]. THG signal is stronger in the body of lesion, indicating that the optical heterogeneity of body of lesion has higher optical heterogeneity compared with sound enamel and surface zone. The above results demonstrate that the multimodal nonlinear optical imaging technique developed is a powerful tool for providing deep insights into the 3-D morphological architectures and biomolecules/biochemical distributions of the tooth with high spatial resolution. The 3-D imaging ability of the multimodal nonlinear optical imaging technique developed could further facilitate early detection and diagnosis of dental caries beneath the enamel surface. One notes that a high NA microscope objective (NA = 1.0) is used in this study for polarized SRS imaging with a high spatial resolution, but a tight light focusing under a high NA objective could result in a generation of z-polarized SRS signals from axially oriented molecules in the samples [26][27][28]. To minimize z-polarization effect, the tooth samples are cut longitudinally for polarized SRS imaging, hence the orientations of crystals predominantly along the enamel rods in enamel [29] are mostly perpendicular to the z component of the linear-polarized excitation fields, making the z-polarized SRS signal being much weaker than SRS signals along the x and y directions [27]. On the other hand, the depolarization ratio of the microscope objective used in this study is measured to be ~0.005, which is much smaller than the difference between the depolarization ratios of the sound enamel and dental caries. Therefore, the contributions of z-polarized SRS signals and the depolarization effect induced by the high NA objective are insignificant, which would not affect the performance of polarization-resolved SRS imaging for optical detection and diagnosis of dental caries.
This work shows that polarization-resolved hyperspectral SRS imaging technique has capability of characterizing the biochemical and biomolecular orientation changes of tooth samples associated with dental caries at submicron scales without labeling. To extend the polarization-resolved hyperspectral SRS imaging technique into in vivo applications in dentistry, some technical advancements on SRS imaging are required. For instance, a miniaturized ultrafast laser system (e.g., fiber-based ultrafast lasers) should be developed serving as pump and Stokes laser beams for tissue excitation. A handheld fiber-optic probe together with a rapid light scanning mechanism should also be designed and developed for accessing to the oral cavity for epi-detected video-rate SRS imaging for real-time tooth diagnosis and characterization.

Conclusions
We demonstrate for the first time that polarization-resolved hyperspectral SRS imaging can be used for rapid characterization and diagnosis of dental caries in a quantitative way without labeling. The polarization-resolved hyperspectral SRS imaging technique provides new insights into the understanding of biochemical and biomolecular changes as well as biomolecular orientation changes associated with carious process, facilitating label-free early detection and diagnosis of carious lesions in dentistry.