Investigation of post-glaucoma-surgery structures by three-dimensional and polarization sensitive anterior eye segment optical coherence tomography ♦

: A sequential case series of post-glaucoma-surgery structures examined by three-dimensional corneal and anterior eye segment optical coherence tomography (3D-CASOCT) and 3D polarization sensitive CASOCT (PS-CASOCT) is presented. A total of 5 patients who underwent glaucoma surgery were included in this study. Of these, 1, 1, and 3 patient underwent trabeculotomy, laser iridotomy, and trabeculectomy respectively. One patient each who had undergone trabeculotomy or laser iridotomy was examined using a prototype 3D-CASOCT. This prototype is based on swept-source OCT technology, uses a probe beam with a center wavelength of 1.31 m m, and has an axial resolution of 11.6 m m and a scanning speed of 20,000 A lines/s. All 3 patients who underwent trabeculectomy were examined by PS-CASOCT, which has similar speciﬁcations to those of 3D-CASOCT, measures the depth-resolved birefringence of a specimen, and yields conventional OCT images. Detailed 3D visualization of the incision site of trabeculotomy and the ablation site of laser iridotomy was achieved using 3D-CASOCT. PS-CASOCT revealed, in addition to the structural details, the birefringent properties of the birefringence a the of ﬁbrosis in PS-CASOCT clinically signiﬁcant for the postoperative assessment of


Introduction
Glaucoma is one of the major causes of blindness; it affects up to 22.5 million people in the world and is responsible for 5.2 million cases of blindness [1,2]. Although the detailed mechanism underlying this disease remains unclear, elevation of intraocular pressure (IOP) is known to play an etiologic role in glaucoma. One of the possible mechanisms of glaucoma development is as follows. Elevated IOP triggers the disease and applies mechanical stress to the biomechanical structures of the optic nerve head (ONH) [3] and/or disturbs ocular circulation. The disorder induced by the mechanical stress and the dysfunction of ocular circulation trigger optic neuropathy.
The only effective strategy for the suppression of glaucoma progression is the reduction of IOP. While pharmacological interventions to reduce IOP are effective in the early stage, surgical intervention, including trabeculectomy, trabeculotomy, and laser iridotomy, are required in advanced or some specific cases. In these surgeries, artificial tissue structures that bypass the drainage of the aqueous humor are created. It is crucial to investigate the morphology of these artificial tissue structures in order to assess their function. The structures include a filtering bleb of trabeculectomy, ablated trabecular meshwork of trabeculectomy, and an abrated hole on the iris of laser iridotomy.
A slit-lamp microscope and an ultrasound biomicroscope (UBM) can be used to observe these structures. However, the slit-lamp microscope cannot reveal the internal morphology of the artificial tissue structure. The UBM on the other hand can reveal the internal structure, but the obtained images are two-dimensional (2D). Furthermore, UBM examination carries a risk of infection since contact with the eye is required. Corneal and anterior segment optical coherence tomography (CAS-OCT) is an emerging modality that can overcome these limitations.
OCT is a noninvasive optical modality that provides cross-sectional images of in vivo tissue at a resolution in the micrometer scale [4]. In the field of ophthalmology, OCT was first applied for retinal imaging [5] and subsequently for the examination of the anterior eye [6,7]. OCT remained a 2D modality until the development of Fourier domain OCT (FD-OCT) techniques, including spectral-domain OCT (SD-OCT) [8][9][10][11] and swept-source OCT (SS-OCT) [12]. This high-speed alternative to conventional OCT (time-domain OCT; TD-OCT) has accelerated the acquisition speed of OCT and enabled the three-dimensional (3D) in vivo examination of human eyes. The first in vivo high-speed 3D anterior segment OCT was performed by the authors in 2005 [13] and was used in several clinical case studies [14][15][16]. Currently, several groups are working to develop different methods for the 3D examination of the anterior eye, and promising results have been presented [17][18][19].
Although 2D and 3D CAS-OCT can be used for the structural assessment of the eye af- ter glaucoma surgery [16,20,55], they do not provide significant information about the tissue property at the surgical site. This is because OCT is only sensitive to backscattering light intensity and not to the other optical properties of tissues. Functional extensions of OCT, such as Doppler OCT [21][22][23][24][25] and molecular-contrast OCT [26,27], as well as post-processing algorithms [28][29][30][31][32] have been studied in order to develop a new type of OCT that can detect multiple optical properties. Polarization -sensitive OCT (PS-OCT) [33][34][35][36], including its Fourier domain version [37][38][39][40][41][42][43][44], has emerged as one of the most promising modalities. PS-OCT provides not only conventional OCT intensity images (I-OCT images) but also a full set or a subset of polarization-sensitive tomographies, including phase-retardation OCT, axis orientation tomography, and diattenuation tomography. When used for the examination of the posterior eye, this modality demonstrates superior performance with regard to assessment of the retinal nerve fiber layer [45,46] and investigation for age-related macular degeneration [47,48]. However, only a few studies have focused on the application of PS-OCT to the anterior eye [49][50][51][52]. This paper aims at providing a reference case series in which tissue structures created during glaucoma surgery were postoperatively examined using non-polarization-sensitive high-speed 3D CAS-OCT (3D-CASOCT) [13] and its polarization-sensitive extension (3D polarizationsensitive CAS-OCT; PS-CASOCT) [52]. One patient each who had undergone trabeculotomy or laser iridotomy was examined using 3D-CASOCT. The filtering blebs in the 3 patients who underwent trabeculectomy were measured using PS-CASOCT.

Three-Dimensional Corneal and Anterior Eye Segment OCT
The postoperative structure of the surgical sites were examined using 3D-CASOCT. Figure 1 shows a schematic of our 3D-CASOCT system. It is based on a standard SS-OCT system. The light source (HSL-2000; Santec, Japan) [53] has a center wavelength of 1.31µm, a scanning bandwidth of 110 nm, a scanning speed of 20 KHz, a lasing duty exceeding 50%, and 3-dB single-path coherence length of 6mm. This spectral property and scanning speed provide a depth resolution of 11.6 µm in air and a measurement speed of 20,000 A lines/s. The light from the light source passes through a fiber Mach-Zehnder interferometer and then illuminates the anterior eye of the patient. A 2D galvanometric scanner scans the probe light according to a raster-scanning protocol. The backscattered probe beam and the reference beam produce spectral interference signals on a balanced photodetector. The interference signal is digitized and is sampled at a sampling frequency of 60 MHz with a digitization resolution of 14 bits. A standard image-reconstruction algorithm for SS-OCT is applied to the digitized spectra, and a 3D tomographic image of the anterior eye is obtained.
A typical 3D OCT volume contains 200 × 200 A-lines, and its acquisition time is 2 s. The typical optical power of the cornea is 5 mW, and sensitivity is 107 dB. The details of this 3D-CASOCT are described in ref. [13]. Figure 2 shows an optical schematic of the PS-CASOCT used in this study. This schematic is similar to that presented in section 2.1, except for the following 3 significant differences.

Polarization Sensitive Corneal and Anterior Eye Segment OCT
The first difference is in the phase-stabilization mechanism. A portion of the probe beam (1%) was trapped and used to illuminate a phase reference mirror. The OCT signal reflected by this reference mirror produces a reference phase, and inter A-line fluctuation of OCT phase is canceled by using this reference phase. This phase stabilization process is mandatory to remove fixed pattern noises originated from the multiple reflections mainly from an electro-optic polarization modulator (EOM) in a source arm and polarization maintaining fiber components in a detection arm of the interferometer. After this phase stabilization, the fixed pattern noises is canceled by a standard average-subtraction algorithm, in which the average of the spectra in an entire B-scan is subtracted from the each individual signal spectrum of the B-scan.
The second difference is the introduction of the EOM in the source arm. The EOM sinusoidally modulates the phase of a single polarization component of the source light, while the orthogonal polarization component is not modulated. The reference arm was configured to sinusoidally modulate the phase of the reference beam in both polarization components. Hence, 2 polarization components are multiplexed in a single spectral interference signal, in which the first polarization component is not altered, while the intensity of the second is modulated.
The third difference is the presence of a polarization-sensitive scheme in the detection arm. The polarization-sensitive detector pair simultaneously acquires 2 spectral interference signals whose polarizations are orthogonal to each other. Since the polarization axis of this polarization-sensitive detector pair is not identical to that of the polarization modulation, each single spectral interference signal contains 2 multiplexed polarization components. The multiplexed polarization components are de-multiplexed by using numerical Fourier transform technique, and finally, 4 polarization-sensitive spectra are obtained.
The 4 OCT signals obtained from this spectral set are fed into a Jones matrix-based PS-OCT algorithm [41,52], and a set of PS-OCT images are obtained, which included cumulative phase-retardation OCT (ϕ-OCT), axis-orientation OCT, and diattenuation OCT images.
A polarization-diversity intensity-OCT (PD-OCT) image, which has similar properties to those of conventional OCT images, was created by averaging the intensities of the 2 OCT images obtained using 2 non-modulated polarization components acquired via 2 detection channels.
Real-time conventional intensity OCT images are reconstructed and displayed during a clinical measurement session, while the polarization sensitive OCT images are reconstructed after closing the measurement session.
It should be noted that the phase-retardation value measured by PS-CASOCT is wrapped and folded within the range of 0 to π radians in the following manner: where φ is the true cumulative phase retardation of the sample, and φ ′ is the phase-retardation value obtained using our PS-CASOCT system. The depth resolution and sensitivity were 11.9 µm in air and 102 dB at 570 µm from the zero-delay point respectively. The power on the cornea was 2.0 mW. Further details regarding this PS-CASOCT can be found in ref. [52].

Subjects
This study included 4 glaucoma patients and 1 patient with primary angle closure. Of the 5 patients, 1, 1, and 3 patient underwent trabeculotomy, laser iridotomy, and trabeculectomy respectively. The details of the subjects are summarized in Table 1.
The examination protocol was designed to adhere to the tenets of the Declaration of Helsinki and was approved by the institutional review boards of the University of Tsukuba and Tokyo Medical University. Written informed consent was obtained prior to 3D-CASOCT and PS-CASOCT imaging.
In this case series, the scanning protocols of OCT depended on the individual subjects. The details of the scanning protocol are summarized in Table 2. The number of A-lines in this table is not necessarily equal to originally obtained OCT volume, because of the truncation of the area-of-interest. Although the volumetric measurements took a few seconds, the fixation of the eyes of the subjects were sufficiently stable, hence no method to cancel the eye motion was applied.

Case 1: 3D-CASOCT of Trabeculotomy
The patient was a 53-year-old Asian woman with a history of secondary steroid induced glaucoma (SIG). The IOP of the right eye was 45 mm Hg. Trabeculotomy was performed in her right eye in Tsukuba University Hospital, and the postoperative IOP was reduced to 13 mm Hg. Measurement with 3D-CASOCT was performed 45 days after the surgery by using a rasterscanning protocol with 200 × 200 A-lines.    Fig. 3(c)). The following properties of this trabeculotomy site were manually measured from this 3D dataset. The site was located 0.23 mm from the angle recess and the size of the ablated site was measured to be 0.17 mm × 0.08 mm. A full 3D dataset is available on-line for detailed observation (23.9 MB). See View 1.

Case 2: 3D-CASOCT of Laser Iridotomy
The patient was a 59-year-old Asian woman with a history of primary angle closure. During consultation at the Tsukuba University hospital, it was found that the anterior chambers of both eyes were shallow, and the IOP in both eyes was 14 mm Hg. Both eyes were classified as Shaffer grade 1, as determined by gonioscopic examination. Nd:YAG laser iridotomy was uneventfully performed in both eyes to prevent acute angle closure crisis. An ablation hole (see appendix C.2) of this patient was observed by a slit lamp examination as shown in Fig. 4(c).
One week after the surgery, the IOPs of the right and left eye were 8 and 9 mm Hg respectively. Her left eye was scanned by 3D-CASOCT. Figures 4(a) and (b) show the horizontal and vertical scans extracted from a single 3D I-OCT volume. Although the laser iridotomy site (white arrow) was very small, it could be easily targeted because of the 3D measurement capability of this CAS-OCT system. Perfect penetration of the iridotomy channel was clearly observed in this OCT volume. The full 3D dataset enables comprehensive observation of the ablation site as shown in Fig. 4(d). This 3D dataset is available on-line (23.9 MB). See View 2. The parameters associated with this laser iridotomy site were manually measured from this 3D OCT volume. The iridotomy hole measured 0.5 mm × 0.8 mm, and the volume removed by iridotomy was estimated to be 0.18 mm 3 .

Case 3: PS-CASOCT of filtering bleb one week after trabeculectomy
The patient was a 69-year-old Asian man with a history of neovascular glaucoma with central vein occlusion in his right eye. He was referred for ophthalmic examination to Kasumigaura Hospital, Tokyo Medical University. He underwent trabeculectomy of the right eye because of increased IOP of 35 mm Hg despite pharmacological therapy. Four days after the surgery, a diffuse filtering bleb (see appendix C.3) was observed on slit-lamp examination ( Fig. 5(a)), and the IOP was decreased to 8 mm Hg.
One week after the trabeculectomy, PS-CASOCT was performed on the filtering bleb. A PD-OCT revealed the internal structure of the filtering bleb as shown in Fig. 5(b). A fluid pool was clearly observed and is indicated by an asterisk (*) in Fig. 5(b). In the ϕ-OCT image shown in Fig. 5(c), marked birefringence was not observed in the conjunctiva, while strong birefringence was observed in the sclera. This contrasting appearance of birefringence is consistent with the finding in normal anterior eye segments.
The 3D interactive observation enables comprehensive and detailed observations of the PD-OCT and ϕ-OCT. These 3D datasets are available on-line. See View 3 for PD-OCT (35.9 MB) and View 4 for ϕ-OCT (35.9 MB).
A set of raw complex OCT volumes (1.0 GB) are also provided, which enables the readers to develop their own PS-OCT algorithm. For the details, see Appendix B.

Case 4: PS-CASOCT of filtering bleb three-year after trabeculectomy
The patient was 66-year-old Asian woman, who originally suffered from adult-onset Still's disease, was medicated with steroids, and consequently developed steroid-induced glaucoma (SIG). She underwent fornix-based trabeculectomy in her right eye in Tsukuba University Hospital. During surgery, 0.04% mitomycin C was instilled into the eye, and a 4 mm × 4 mm scleral flap was created. The postoperative IOP was maintained at 18 mm Hg on treatment with topical latanoprost, while the preoperative IOP was 44 mm Hg.
A diffuse filtering bleb was observed at the 11 o'clock position on slit-lamp observation. Three year after the surgery, the filtering bleb was examined by non-polarization sensitive 3D-CASOCT. The detail of this 3D structural investigation was previously published (the second case of ref. [16]).
The same region was scanned by 3D PS-CASOCT 1 year after the first 3D-CASOCT examination (4 years after the surgery). The preceded slit-lamp examination did not reveal the internal structure of the bleb because of severe scarring of the conjunctiva (Fig. 6(a)). The PD-OCT image clearly depicted the fluid pool in the filtering bleb, as shown in Fig. 6(b) (white asterisk (*)). In the ϕ-OCT image shown in Fig. 6(c), moderate birefringence was observed around the anterior inner wall of the fluid pool. Since the normal conjunctiva is known not to exhibit birefringence, the observed birefringence may indicate fibrosis in this region, although further intensive studies are required for a more reliable conclusion.
The 3D interactive observation enables comprehensive and detailed observations of the PD-OCT and ϕ-OCT. These 3D datasets are available on-line. See View 5 for PD-OCT (35.9 MB) and View 6 for ϕ-OCT (35.9 MB).
A set of raw complex OCT volumes (1.0 GB) are also provided, which enables the readers to develop their own PS-OCT algorithm. For the details, see Appendix B.

Case 5: PS-CASOCT of filtering bleb four-year after trabeculectomy
The patient was a 50-year-old Asian woman with a history of primary open-angle glaucoma in the right eye, and she was referred for ophthalmic examination to Kasumigaura Hospital, Tokyo Medical University. She underwent trabeculectomy in her right eye because of elevated IOP of 40 mm Hg, despite the administration of maximally tolerated pharmacological therapy.
Four years after the surgery, a non-functioning filtering bleb was observed on slit-lamp examination, as shown in Fig. 7(a). PD-OCT revealed the internal structure of the bleb. Several small fluid pools are observed as shown in Fig. 7(b) (the red arrows). The remnant of the fluid pool is also visible in the same image (green arrow). It is noteworthy that this region exhibited moderate birefringence, as observed on ϕ-OCT as shown in Fig. 7(c) (green arrow). This moderate birefringence may indicate abnormal fibrosis in this region.
The 3D interactive observation enables comprehensive and detailed observations of the PD-OCT and ϕ-OCT. These 3D datasets are available on-line. See View 7 for PD-OCT (35.9 MB) and View 8 for ϕ-OCT (35.9 MB).
A set of raw complex OCT volumes (1.0 GB) are also provided, which enables the readers to develop their own PS-OCT algorithm. For the details, see Appendix B.

Discussions and Conclusion
In patient 4, the area of the fluid pool was manually segmented by an experienced ophthalmologist by using a series of 2D cross-sectional tomographic images [16]. Although the segmented volume provided several clinically significant parameters, this process is time consuming and not applicable in the clinical setting. The development of an automated algorithm for the segmentation of the filtering bleb is an important future issue. The following engineering-related issues are expected during the automation process.
The signal strength of FD-OCT depends on the depth of the signal; i.e., the signal strength decays as a function of the distance from the zero-delay point of the OCT interferometer [10,54]. This signal decay results in a systematic offset of the signal strength. Further, the signal strength and the image contrast of a tissue are strongly affected by the scattering properties of the superficial tissues. The development of a segmentation algorithm that is robust to these systematic and unsystematic signal perturbations is the next issue that should be addressed.
The contrast of the filtering bleb is related to the optical properties of the bleb tissues [55], and it varies from patient to patient. This difference in the contrast of the filtering bleb should be further investigated from a clinical perspective. Further knowledge of the OCT contrast of the filtering bleb will lead to improvements in the automatic segmentation algorithm.
The size and position of the incision or ablation site is important for trabeculotomy and laser iridotomy. For further analysis, it is crucial to automatically determine the shape of an incision site and the region of an ablation site.
Bullous keratopathy is an infrequent but serious complication of laser iridotomy [56,57]. Alteration of the fluid-dynamic properties of aqueous humor after laser iridotomy is one of the suspected reasons for this complication [58]. The development of an accurate and automated segmentation algorithm will enable the creation of a complete 3D numerical model of the ablation site. This numerical model will in turn enable detailed analysis of the mechanical and fluid dynamic functions of the ablation site.
The ϕ-OCT images of subjects 3, 4, and 5 showed a variety of contrasts of the tissues around the filtering bleb. In contrast to ϕ-OCT, PD-OCT does not clearly differentiate among patients. Large variations in the contrast of ϕ-OCT may be related to the individual differences among subjects and variations in the interval between surgery and the performance of postoperative OCT. For example, it is known that fibrosis occasionally occurs in old filtering blebs. On the basis of the examination findings in the posterior eye segment, fibrosing tissue is selectively visualized by PS-OCT [48]. Such tissue properties are known to be related to the functions of the filtering bleb; hence, the assessment of tissue properties is crucial for the follow up of trabeculectomy. PS-CASOCT will enable routine noninvasive assessment of the filtering bleb.
Other than the evaluation of glaucoma surgeries, PS-CASOCT has more potential applications. Since PS-CASOCT visualizes trabecular meshwork with higher image-contrast than conventional CAS-OCT [49,52], PS-CASOCT may be useful for the screening of angle closure glaucoma. Cornea consists of collagen and is known to be birefringent. This birefringent property may be useful for the evaluation of the abnormality of cornea, e.g. keratoconus [51].
In conclusion, 3 filtering blebs after trabeculectomy, and an incision site of trabeculotomy and an ablation site of laser iridotomy were examined by 3D-CASOCT and PS-CASOCT. The case series presented here is of clinical importance. The development of an automated quantitative image analysis technique will in the future provide results of greater clinical significance.

A. Appendix: Normal Anterior Chamber Angle
The anterior chamber angle (ACA) of a normal subject was examined using PS-CASOCT after obtaining written informed consent. The subject was a 48-year-old Asian man; no marked abnormality was present in his anterior eye. The details of the subject and the scanning protocol are presented in Tables 1 and 2 as subject 6. Although this result has already been presented in ref. [52], this appendix aims at providing a full 3D dataset of PD-OCT and ϕ-OCT as a reference standard of PS-CASOCT. Figure 8 has been reprinted from ref. [52] and shows (a) PD-OCT, (b) ϕ-OCT, and (c) axisorientation OCT images of the ACA. In the PD-OCT image (Fig. 8(a)), a trabecular meshwork (TM) appears as a diffuse hyper scattering region (indicated by an arrow). In contrast, the TM appears as a winding band in the ϕ-OCT. In the axis-orientation image, the variance of the signal decreases at the TM because the TM shows strong birefringence, which results in a high signal-to-noise ratio. The 3D datasets are available on-line. See View 9 for PD-OCT (65.6 MB), View 10 for ϕ-OCT (65.6 MB), and View 11 for axis-orientation OCT (65.6 MB). A set of raw complex OCT volumes (1.9 GB) are also provided, which enables the readers to develop their own PS-OCT algorithm. For the details, see Appendix B.

B.1. Data Format
One of the purposes of this paper is to provide a reference dataset of PS-OCT to developers and researchers who wish to develop their own PS-OCT algorithm. For this purpose, we provide raw complex OCT volumes of cases 3, 4, 5 and 6 as Interactive Science Publishing (ISP) datasets of Case0302 (case 3), Case0402 (case 4), Case0502 (case 5) and Case0603 (case 6). A set of OCT volumes of each case consists of 4 complex OCT volumes labeled as H0.csg, H1.csg, V0.csg, and V1.csg. Where H and V represent that the datasets were respectively acquired by horizontal and a vertical detection channels of PS-OCT, while 0 and 1 indicate whether this volume is reconstructed from a non-intensity-modulated spectrum (0) or an intensity-modulated spectrum (1). For the details of these detection channels and the modulation, refer our previously published paper [52].
In these raw data volumes, numeric values are represented by a single precision complex (CSG, cmplx64) format of LabVIEW (National Instruments, TX), which is defined as the following code written in C. The data are stored with big-endian byte order. The dimensions of the volumes are as summarized in Table 2. The data position of a point (x, y, z) in the volume file is calculated as Data position = (∆x × ∆z × y + ∆z × x + z) × 8 bytes, where the data position is relative to the starting point of the file. ∆x, ∆y and ∆z are the numbers of voxels along horizontal, vertical and depth directions, respectively, and x, y, and z are voxle indexes along the corresponding directions. The origin is set to nasal-superior-anterior corner of the OCT volume. The examples of LabVIEW programs to read the OCT data are shown in Fig. 9. Figure 9(a) is a program which reads a complex value at a single point of an OCT volume, while Fig. 9(b) is for reading a single horizontal scan from the OCT volume.