2.1.

Specimen Preparation

The imaging was performed at the Oral Medicine Clinic, Sir Charles Gairdner Hospital (Perth, Western Australia). The study was approved by the Human Research Ethics Office of the University of Western Australia (UWA Ethics number RA/4/1/8562) and written informed consent was obtained from all participants. Over a 3-day period, we measured 12 freshly excised tissue samples from different locations in the oral cavity and covering a range of pathologies (see Table 1 for the clinical presentation of excised samples). Excised samples were transferred from the surgery to a nearby office and imaged with PS-OCT within 5 min of excision. Following the PS-OCT measurement, samples were immersed in a fixative solution (10% neutral-buffered formalin) and stored at room temperature. The measurement of all stored samples was repeated at 24 h from excision ($\pm 45\min$). In routine pathology processing of ex vivo oral cavity tissue, samples of the size used in this study undergo 24-h fixation prior to further processing; therefore, we assumed 24 h was sufficient elapsed time for samples to be fully fixed. For reference, Hsiung et al.²⁴ reported 18 h as the time necessary to ensure complete tissue fixation for hamster cheek pouch samples. Additionally, for further confirmation of fixation status, five samples were also measured at 48 h after excision.

Table 1

Ex vivo samples used for the study described by site in the oral cavity and clinical presentation.

2.2.

Polarization-sensitive OCT

Imaging was performed with a custom PS-OCT scanner developed in-house and used previously for needle-based measurements of breast samples¹³ and adapted recently for bulk optics-based measurements.²⁴ The technical specifications of the scanner are summarized as follows. The system operated with an A-line repetition rate of 50 kHz at the center wavelength of 1310 nm and a scanning range of 110 nm. We employed passively depth-encoded polarization multiplexing and polarization-diverse detection to measure the full Jones matrix containing the cumulative polarization information of backscattered light after round-trip propagation through the sample.²⁶ To provide sufficient imaging range for depth-encoded multiplexing, a custom-made sampling clock frequency-doubling circuit was used. The resolution of the system was $13\mu \mathrm{m}$ and $20\mu \mathrm{m}$ in tissue in the axial (assuming a tissue refractive index $n=1.4$²⁷) and lateral directions, respectively. Single volumetric data sets consisting of $1000\times 1000$ A-scans covering an area of $5.1\times 5.1\mathrm{mm}$ with a depth range of 3.5 mm in air were captured within 20 s for each sample. Further technical details on the instrument (not relevant to the current study) can be found in Refs. 13 and 24. The $XY$ scan area was sufficient to image the whole sample in the lateral dimensions.

Data processing to extract local birefringence is summarized as follows. Mueller–Jones matrices were constructed from the measured Jones matrices and spatially averaged before extraction of the local birefringence using a differential Mueller matrix algorithm.¹³ In contrast to the majority of previous studies, where average birefringence is computed from the cumulative retardance from the tissue surface to a given depth, here, we provide local birefringence at close to the same resolution as the OCT intensity images. Separately, in our recent in vivo study,¹⁷ we have shown that results presented by means of cumulative retardation provide lower contrast, especially at greater depths, compared with local birefringence. An averaging kernel of $26\mu \mathrm{m}$ axially and $40\mu \mathrm{m}$ laterally was used to improve the local birefringence contrast at the expense of a twofold reduction in spatial resolution. Both OCT intensity and local birefringence cross-sectional images are presented in grayscale in Figs. 1(a) and 1(b). Better insight regarding the features of measured samples is provided using an en face view [Figs. 1(c) and 1(d)]. The measured image data were first flattened by automatic subtraction of the extracted position of the tissue surface, so each en face slice corresponds to a given depth below the tissue surface (as marked by the red dotted curve on the cross-sections in Figs. 1(a) and 1(b). We note that, for some of the samples, the surface of the excised tissue does not correspond to the natural anatomical surface of the tissue before extraction. The flattening procedure was introduced to create en face slices and to enable automation of the quantitative analysis.

Fig. 1

Cross-sectional (B-scan) PS-OCT images of the ex vivo right ventral tongue sample (sample #1 from Table 1): (a) conventional OCT intensity image, (b) depth-resolved birefringence ($\mathrm{\Delta }n$) revealing increased contrast compared to OCT intensity image, (c) en face slices from the depth indicated by the red dotted curve in (a), and (d) depth-resolved birefringence from the depth indicated by the red dotted curve in (b). Yellow dotted lines correspond to the position of cross-sectional images. For images on (b)–(c), regions corresponding to low SNR, as measured in the conventional OCT images, were masked out. Images were scaled in depth assuming a refractive index of $n=1.4$. Scale bar: 1 mm.

Surface flattening was performed and en face slices were prepared for all measurements. The orientation angle (in the $XY$ plane) of the same sample was found to differ between measurements for some samples. This was irrelevant for quantitative analysis of image contrast. However, for visual presentation and comparison purposes, the orientation angle difference was manually removed in postprocessing. The angle registration was performed by visual inspection based on sample topography features (see Figs. 1Fig. 2–3).

Fig. 2

Paired—OCT intensity on the left and birefringence on the right—en face slices of an ex vivo right ventral tongue sample (sample #1 from Table 1) at various depths (a)–(c) for freshly excised tissue, and 24 h, and 48 h after fixation was initiated. Scale bar: 1 mm. The yellow asterisk indicates an area where birefringence contrast is locally reduced. The OCT intensity and birefringence scale bars are the same as in Fig. 1 and omitted here for clarity.

Fig. 3

Paired—OCT intensity on the left and birefringence on the right—en face slices of the ex vivo right soft palate sample (sample #3 from Table 1) at various depths (a)–(c) for freshly excised tissue as well as 24 h and 48 h after fixation was initiated. The yellow and orange dashed-line boxes indicate the regions further visualized and analyzed in (d)–(k). Scale bar: 1 mm. Abbreviations used in (g) and (k): LB, left bottom and RT, right top related to the yellow dotted line on zoomed birefringence maps (d)–(f), LT, left top and RB, right bottom for similar orange dotted line (h)–(j). The OCT intensity and birefringence scale bars are the same as in Fig. 1 and omitted here for clarity.

We automated the quantitative analysis of local birefringence contrast, thereby avoiding observer bias. We selected data for analysis using a binary mask (separate masks were created for each en face depth slice and each time point) by thresholding of OCT intensity data followed by two morphological operations: (1) closing of the image to fill the gaps in the mask and (2) erosion of the image to remove bright pixels not corresponding to the mask, applied to the mask to eliminate noise and irrelevant artifacts. The intensity-based masks were used on the corresponding en face slices of local birefringence. All procedures were optimized on one randomly selected data set, visually tested on three additional data sets, and then applied in unsupervised processing of all data sets. The contrast was calculated for each en face slice, starting from the surface to a depth of almost $800\mu \mathrm{m}$, determined as the ratio of the standard deviation (${\sigma }_{\mathrm{\Delta }n}$) to mean value of birefringence ($\overline{\mathrm{\Delta }n}$) (after application of the averaging kernel):

For samples #9 and #11, the analysis depth was reduced to 600 and $450\mu \mathrm{m}$, respectively, as no sample-related signal was present below these depths.

For each sample and time point, we performed a statistical comparison for every en face slice ($N=112$) between measurements of freshly excised samples and fixed samples measured 24 h (and 48 h where available) after the commencement of fixation. The statistical comparisons were first tested for normality of the distribution using the Shapiro–Wilk test.²⁸ As all pairs were non-normally distributed, the Wilcoxon test²⁹ with 5% significance level was used. Additionally, we analyzed the contrast differences at given depths in the complete sample population. For every depth slice, we calculated the contrast differences between the fresh and 24-h time points as well as between the fresh and 48-h time points. The mean difference and standard deviation for each slice were calculated and plotted versus depth.

3.1.

Local Birefringence Qualitative Contrast

Figure 2 shows en face OCT intensity and local birefringence sections extracted from different depths below the sample surface at the different time points of the fixation process. The sample was excised from the right ventral tongue and was preassessed as squamous papilloma. For this sample, we observe high variability of birefringent structures, which most likely represent highly aligned collagen fibers in the reticular layer, as described in our previous study.²¹ In general, in Fig. 2, we observe the contrast in both intensity and birefringence due to tissue alteration induced by the fixation procedure. For the deepest image sections [Fig. 2(c)], some examples of decreased contrast are observed (e.g., at the location marked by the yellow asterisk). This may be due to the effect of axial tissue shrinkage during fixation causing previously deeper layers to be shifted closer to the surface. The overall and rigorous assessment of the birefringence contrast change, however, is difficult based on visual inspection alone; therefore, we undertook the quantitative analysis described above and presented in the following sections.

The second example [Figs. 3(a)–3(c)] taken from the right soft palate reveals changes in the pattern of birefringent features. Magnified views [Figs. 3(d)–3(j)] depict visible improvement in the birefringence contrast during the fixation process. Two regions (yellow and orange boxes in Figs. 3(a) and 3(b) were selected to show the effect quantitatively. Zoomed maps of birefringence for freshly excised [Figs. 3(d) and 3(h)], 24 h [Figs. 3(e) and 3(i)], and 48 h [Fig. 3(f) and 3(j)] after fixation, as well as plots of birefringence extracted along selected lines [Figs. 3(g) and 3(k)], are presented. We chose to present data from the soft palate sample (Fig. 3) at different depths compared to the ventral tongue sample (Fig. 2) to help better appreciation of the variations in sample birefringence due to fixation, including with depth.

3.2.

Local Birefringence Quantitative Contrast

To provide more than the visual assessment of the effects of tissue fixation on polarization contrast, as described in Sec. 2, we applied automatic contrast analysis for both intensity and birefringence images. The results are presented as a boxplot (Fig. 4). For further statistical assessment, the paired $t$-test was applied to the results. Statistically significant differences were found for all cases except when fresh tissue was compared with 24 h after fixation for samples #1, #2, and #9.

Fig. 4

Quantitative comparison of contrast change during the fixation process for all measured samples (green color represents fresh; red color, 24 h after fixation was initiated; blue color, 48 h after fixation was initiated). Red crosses represent outliers (values more than 3 times the interquartile range away from the bottom or top of the box). Asterisks indicate statistically significant differences in 11 out of the 12 samples.

To provide more insight on how the contrast changes versus depth in the sample, we performed statistical (mean and standard deviation) slice-by-slice assessment of the differences in contrast between freshly measured samples and those measured 24 or 48 h after fixation. At each sample depth, the contrast change (24 h versus fresh and 48 versus fresh) was calculated. Depth by depth, we plot the mean contrast change over all samples with the ribbon representing standard deviation (Fig. 5). On average, we observe up to 10% increase in mean contrast to a depth of $500\mu \mathrm{m}$ followed by up to 20% increase for deeper layers. The variation, presented as a ribbon plot, is higher at deeper layers, where the SNR is usually lower than for preceding layers.

Fig. 5

Change in fixed-sample contrast relative to fresh-sample contrast versus depth. Mean contrast change versus depth is presented with red-dashed curve for changes between fresh sample and 24 h after fixation, and blue-dotted curve for changes between fresh sample and 48 h after fixation. Corresponding ribbon plots represent ±1 standard deviation.