«Novel Biophotonic Imaging Techniques for Assessing Women’s Reproductive Health by Tyler Kaine Drake Department of Biomedical Engineering Duke ...»
Chapter 6 presents on a pilot clinical study focusing on a different aspect of women’s reproductive health.
A clinical a/LCI instrument is used in a pilot study of ex vivo cervical tissue to assess the viability of using a/LCI to identify cervical dysplasia.
Angle- resolved low coherence interferometry (a/LCI) is an optical biopsy technique used to measure the average size and optical density of cell nuclei as a function of depth within epithelial tissue in order to assess tissue health.
The angular distribution of elastically scattered light from cell nuclei and other small scatterers in the localized layers of targeted tissue is collected and compared to Mie Theory.
a/LCI obtains depth- resolved nuclear morphology measurements without the use of exogenous contrast agents, with submicron accuracy.
Section 6.4 describes the study design of the pilot ex vivo cervical dysplasia study in which the a/LCI device was used tissue sample drawn from patients undergoing cervical come biopsies or hysterectomy procedures.
Twenty- three biopsy sites from 20 tissue specimens were scanned in this study, and nuclear morphology results at the basal layer (200 µ;m deep) were compared to co- registered histopathological diagnosis.
95 An additional finding of this study was found that a/LCI A- scans (depth- scans) were able to identify endocervical tissue from ectocervical tissue based on the scan intensity with depth.
This pilot study demonstrated the feasibility of a/LCI to detect cervical dysplasia using ex vivo tissue.
This study concludes that the average nuclear diameter in the basal layer of ectocervical epithelium showed a statistically significant increase in size for dysplastic tissue compared with non- dysplastic tissue types, a finding which is consistent with a/LCI results from other tissue types.36, 37, 82
6.2 a/LCI clinical instrument The Fourier- domain a/LCI system has been effective in identifying dysplasia in both human and animal epithelial tissues with high sensitivity and specificity.34- 40 Here, the significant operational features of the a/LCI device are described.
The a/LCI instrument is based off of a modified Mach- Zehnder interferometer, and the optical layout is shown in Figure 6.1.
96 Figure 6.1:
Diagram of the clinical a/LCI instrument, as taken from Zhu et al.83 Part (a) shows the optical layout, part (b) is a close up of the probe tip while (c) is a photograph of the probe face.
Like mLCI, light from a SLD (λ0 = 830 nm) is split into two arms, a reference and sample.
The fiber optic splitter allows 95% of the light to be coupled into the sample arm where a polarization controller is used to optimize power through an inline fiber polarizer.
The light output from the polarizer is p- polarized and is delivered to the sample via polarization maintaining (PM) fiber.
Polarization control in the sample arm of the a/LCI instrument is critical because the angular scattering profile from the sample depends on the polarization configuration of the delivered light.
Light scattered from a sample is collected by a coherent fiber bundle and then mixed with a collimated reference field.
The combined field is detected by an imaging spectrometer.
The probe tip assembly is shown in Figure 6.1(b).
As mentioned above, a PM fiber delivers p- polarized light, and this light is collimated by a miniature drum lens before it is incident upon the sample.
Scattered light is then collected by the drum lens 97 and focused onto the face of the coherent fiber bundle.
The drum lens acts to convert angularly scattered light into a spatial distribution across the face of the bundle.
1 coverglass acts as custom shaped to serve as an optical window of the probe tip, and the entire assembly is protected by a lens tube sealed inside a short section of Teflon (PTFE).
This assembly is coupled to a section of PEEK tubing which protects the length of the fiber bundle.
The collected light is spatially filtered by the entrance slit of the spectrometer, to allow just a small central section, as shown in Figure 6.1(c), to be detected.
Photographs of the clinical a/LCI system.
Part (a) shows the system as used in a clinical setting, part (b) shows the optical components with the cover removed, and part (c) shows the fiber probe tip next to a US dime.
(SLD- superluminescent diode;
SP- imaging spectrometer;
L1, L2, L3- lenses;
SR- strain relief;
FS- fiber splitter;
MF- matching fiber;
PC- polarization controller).
Image taken from Zhu et al.83 The complete a/LCI system is shown in Figure 6.2(a), as mounted onto a 27 inx18 in.
utility cart, in preparation for clinical data collection.
The optical components are mounted into an optical breadboard, as seen in Figure 6.2(b), and shielded with an 98 aluminum cover, which protects critical optical components.
During data collection, the probe shown in Figure 6.2(c) is held on the extension tray, when not being used to scan tissue.
Complete characterization of the a/LCI instrument, including performance of the probe, optical properties, and phantom measurements, are described in detail by Zhu et al.83
6.3 a/LCI data acquisition and processing In order to collect light- scattering data from ex vivo epithelial tissue, the a/LCI fiber probe is manipulated by hand and placed in contact with the tissue surface.
Gentle pressure is applied to create a flat tissue surface and to slightly stretch the tissue across the optical window of the probe, creating repeatable measurement geometry.
Once the probe is in place, twenty signal acquisitions are taken at the optical biopsy site.
The probe is then removed, and the site is marked with India Ink for pathological co- location.
For each tissue specimen in the study described in Section 6.4, this was repeated 2- 4 times at sites of suspected dysplasia for conization samples, or randomly on the ectocervix for hysterectomy samples.
Once captured, the signal at each pixel in the 2- dimensional CCD array is a combination of the signal and reference fields (Es, Er), and relates to a specific combination of wavelength and scattering angle (λm, θn).
Equation (6.1) describes the
where (m, n) correspond to a pixel in the CCD, ϕ is the phase difference between the fields and denotes a time average34.
First, the average intensities of the sample and reference beam are subtracted to remove background signal.
Next, the spectrum is converted from wavelength to wavenumber using a cubic spline interpolation, and chromatic dispersion is digitally corrected, similar to steps in mLCI processing as described in Section 3.2.1.
At each scattering angle, a 1- D Fourier transform is performed to generate an A- scan, or depth- scan, of the sample.
A sample 2- D data array for scattering by a calibration sample is shown in Figure 6.3(a).
Typical a/LCI data.
Part (a) shows the 2- D data array where depth (in OPL) is shown on the x- axis and scattering angle is shown on the y- axis for light collected from a scattering sample.
Part (b) shows a single A- scan which has been created by binning across angle.
The data in (b) are an example of an angle integrated A- scan for squamous epithelium from the esophagus.
The gray area shows an region of interest where scattering profiles are compared with Mie Theory.
Part(b) of figure is taken from Zhu et al.83 100 For cell nuclei size measurements, the image is then binned into 50 µ;m depth segments, and the angular scattering profile is analyzed for each depth bin, as shown by the grey lines in Figure 6.3(b).
Scattering profiles are compared to a range of profiles in a Mie theory generated database, chi- squared values are calculated for each comparison, and the minimum chi- squared Mie theory database size match is reported as the mean nuclear size for that tissue depth.
From the ectocervical tissue examined in this study, the basal layer was seen in the pathology slides to exist approximately 200- 250 µ;m deep in the tissue, and thus our analysis focuses on so the scattering angles at this depth.
6.4 Study design Angle- resolved LCI (a/LCI) has been shown to provide depth resolved quantitative nuclear morphology data that can be used to assess the health of epithelial tissues.
Previous studies, performed over the last 10 years of a/LCI development, have found that the technique is capable of identifying dysplasia in epithelial tissue with high sensitivity and specificity.34- 40 Here, a pilot clinical study is described which was performed to assess the ability of a/LCI in detecting cervical dysplasia, or CIN, in ex vivo tissues.
The clinical platform used for this study has been thoroughly described in Section 6.2 of this dissertation.
The pilot a/LCI study was performed in the Duke University Medical Center pathology lab on tissues drawn from 20 participants undergoing hysterectomy or cervical cone biopsy surgeries.
Immediately after tissue was resected 101 via surgery, it was taken to the pathology lab and placed at a dissection station.
The a/LCI probe was then used to scan 2- 4 optical biopsy sites on each tissue sample, as guided by the surgeon or attending pathologist.
At each measurement point, 20 acquisitions were taken.
Optical biopsy locations were marked with India ink and the tissue was fixed in formalin.
Tissues from 9 hysterectomies and 11 cervical conizations were scanned in the trial, for a total of 23 examined sites.
Histological sections of these tissues were prepared as part of routine gynecological care and analyzed by a pathologist to determine disease states.
The diagnoses for the 23 paired sites were then compared retrospectively with a/LCI nuclear morphology measurements.
While each tissue was scanned with a/LCI at least two sites, not all scanned points were identifiable in the histology sections.
In addition, data from two scanned sites were found to have been located on endocervical tissue upon pathological evaluation.
These results were omitted from the analyses.
Finally, one scanned site was found to have an epithelial layer of 600 µ;m thick, which prevented identification of the basal layer in the a/LCI scan.
This data point was also omitted from the analysis.
6.5 Data analysis The a/LCI were processed in a manner as described in Section 6.3 and in greater detail by Brown et al.34 Concisely, the data were processed to isolate the scattering components from cell nuclei in 50 µ;m depth segments.
The basal layer, the area of interest, ranged about 200- 250 µ;m deep in the cervical epithelium so the nuclear diameter and nuclear 102 density were analyzed at this depth.
In the analysis, nuclear density is the relative ratio of the average refractive index of the cell nuclei to the surrounding cytoplasm.
Each scan was examined to ensure that there was sufficient signal as well as to identify the surface layer of the tissue.
The measured nuclear morphological characteristics were compared retrospectively to pathological classification of co- registered biopsy sites, as determined by a trained pathologist.
6.6 Nuclear morphology results Of the 23 paired sites that were examined, 16 sites were determined to be normal ectocervical tissue by pathological evaluation.
Metaplasia was found at 4 tissue sites, and 3 were determined to be dysplastic.
The average nuclear morphology measurements at each of the 50 µ;m depth segments for the tissue types are shown in Table 6.1.
Average nuclear morphology measurements at each of the 50 µ;m depth segments for optical biopsies as measured by a/LCI.
Statistically significant differences in nuclear size were found at the 200 µ;m depth (p = 0.002) when dysplastic tissues were compared with normal tissue.
When comparing metaplastic tissue to dysplastic tissue, there was a statistically significant difference in nuclear density at 100 µ;m depth (p = 0.033).
Average nuclear diameter at the basal layer (200 µ;m depth) is plotted for the three tissue diagnosis types in Figure 6.4.
Normal tissue had an average nuclear diameter of 9.17 ± 1.24 µ;m (mean ± std dev), metaplastic tissue had an average nuclear diameter of 9.75 µ;m ± 1.44 µ;m, and dysplastic tissue had an average nuclear diameter of 11.83 ± 0.07 µ;m.
8.0 6.0 4.0 2.0
To further investigate the relationship between sensitivity and specificity for the data, receiver operating characteristic (ROC) curves were created using the statistically significant morphological measurements as discriminates.
The presence of dysplasia was used as a binary classifier;
with dysplasia being the positive discriminate.
ROC curves for nuclear diameter of normal and dysplastic tissue at the basal layer is shown in Figure 6.5(a).
An ROC curve for nuclear density at 100 µ;m depth for metaplastic and dysplastic tissue is shown in Figure 6.5(b).
Nuclear diameter at the basal layer proves a good predictor for the presence of dysplasia with an area under the curve (AUC) of 0.92.
Likewise, nuclear density at 100 µ;m depth classified dysplastic tissue from metaplastic successfully, with an AUC of 0.96.
(a) Nuclear diameter for normal and dysplastic tissue at the basal layer and (b) nuclear density for metaplastic and dysplastic tissue at 100 µ;m depth.
A scatter plot of average nuclear diameter and nuclear density was also created for the cervical epithelium basal layer, and is shown in Figure 6.6.