The authors have declared that no competing interests exist.
Conceived and designed the experiments: HSH JGS BHL TJE CKJ. Performed the experiments: HSH JGS BHL TJE CKJ. Analyzed the data: HSH JGS BHL TJE CKJ. Contributed reagents/materials/analysis tools: HSH JGS BHL TJE CKJ. Wrote the paper: HSH JGS BHL TJE CKJ.
Recently, we reported obtaining tomograms of meibomian glands from healthy volunteers using commercial anterior segment optical coherence tomography (AS-OCT), which is widely employed in clinics for examination of the anterior segment. However, we could not create 3D images of the meibomian glands, because the commercial OCT does not have a 3D reconstruction function. In this study we report the creation of 3D images of the meibomian glands by reconstructing the tomograms of these glands using high speed Fourier-Domain OCT (FD-OCT) developed in our laboratory. This research was jointly undertaken at the Department of Ophthalmology, Seoul St. Mary's Hospital (Seoul, Korea) and the Advanced Photonics Research Institute of Gwangju Institute of Science and Technology (Gwangju, Korea) with two healthy volunteers and seven patients with meibomian gland dysfunction. A real time imaging FD-OCT system based on a high-speed wavelength swept laser was developed that had a spectral bandwidth of 100 nm at the 1310 nm center wavelength. The axial resolution was 5 µm and the lateral resolution was 13 µm in air. Using this device, the meibomian glands of nine subjects were examined. A series of tomograms from the upper eyelid measuring 5 mm (from left to right, B-scan) × 2 mm (from upper part to lower part, C-scan) were collected. Three-D images of the meibomian glands were then reconstructed using 3D “data visualization, analysis, and modeling software”. Established infrared meibography was also performed for comparison. The 3D images of healthy subjects clearly showed the meibomian glands, which looked similar to bunches of grapes. These results were consistent with previous infrared meibography results. The meibomian glands were parallel to each other, and the saccular acini were clearly visible. Here we report the successful production of 3D images of human meibomian glands by reconstructing tomograms of these glands with high speed FD-OCT.
Dry eye, or keratoconjunctivitis sicca, is defined as “a disorder of the tear film due to tear deficiency or excessive evaporation that causes damage to the interpalpebral ocular surface and is associated with symptoms of discomfort”
The meibomian glands are modified sebaceous glands in the eyelid that secrete lipids onto the preocular tear film
Recently, we reported obtaining tomograms of meibomian glands from healthy volunteers using commercial anterior segment optical coherence tomography (AS-OCT) (Visante, Carl Zeiss Meditec, Dublin, CA, USA), which is widely employed in clinics for the anterior segment (including the cornea and anterior chamber angle)
We can create a 3D skull image by reconstruction of the serial axial tomograms of the skull computed tomography (CT). This is a typical example of 3D reconstruction of medical images in hospitals. In the same manner, we could produce a 3D meibomian gland image by reconstruction of serial OCT images of the gland. However, the creation of 3D images was not possible with AS-OCT, because it does not possess a 3D reconstruction function. Thus, in this study, we report the creation of 3D images of meibomian glands by reconstructing the tomograms of these glands using high speed Fourier-Domain OCT (FD-OCT) developed in our laboratory.
This research was carried out at the Department of Ophthalmology, Seoul St. Mary's Hospital (Seoul, Korea) and the Advanced Photonics Research Institute of Gwangju Institute of Science and Technology (Gwangju, Korea) with two healthy volunteers and seven patients with MGD in November, 2011. The study followed the principles of the Declaration of Helsinki, and the Institutional Review Board of Seoul St. Mary's Hospital approved the study. The purpose of this study and methods were explained to the subjects and written consent was obtained from them. The two healthy volunteers were: subject A, a 35 year-old man; subject B, a 30 year-old man. Subjects A and B were initially examined by slit-lamp biomicroscopy to confirm the absence of abnormalities of the eyelid and ocular surface. We used slit-lamp biomicroscopy and noncontact meibography to diagnose MGD. We examined the lid margin for thickening, erythema, hyperkeratinization, vascularization, telangiectasia, notching or orifice capping. We applied pressure to the lid margin to observe the nature of the secretions. Corneal and conjunctival staining was performed with fluorescein or rose Bengal dye. Finally, noncontract meibography was performed and the change in the meibomian gland was scored using the meiboscore system.
First, we took an infrared photograph of the meibomian gland using the infrared meibography system modified from noncontact infrared meibography
We developed a real time imaging FD-OCT system based on a high-speed wavelength swept laser shown in
The axial resolution was 5 µm and the lateral resolution was 13 µm in air. Interference signal data from the sample scanner for the human meibomian gland and the reference arm were acquired and transferred to a polarization controller (PC) using a high speed digitizer (PX14400, Signatec, Lockport, IL, USA) with 425 × 106 samples/second and 14 bit resolution. For each A-line, 4096 data points were used to reduce the sensitivity roll-off from the data re-sampling and the interpolation processing. We enhanced the quality of the image by repeating the A-scans twice for each point and averaging them. Then, we performed the B-scan (500 A-scans) and C-scan (200 B-scans). The time required for a 2D image (B-scan, 700×500 pixels) was 21 ms and that for the 3D imaging (C-scan, 700 × 500 × 200 voxels) was 4.2 s (21 ms × 200) irrespective of the subject. A700 pixel height was equivalent to an optical distance of 4.81 mm in air.
The following procedure was used for the 3D OCT meibography. The subject sat in front of the scanner and placed his or her face on a headrest (
After completing FD-OCT scanning, the 3D image of the meibomian glands was reconstructed using 3D “data visualization, analysis, and modeling software” (AMIRA software; Mercury Computer Systems, Chelmsford, MA, USA). We cropped the volume data to remove the image of the palpebral conjunctiva and to highlight the region of the meibomian glands. We applied the following image processing protocol to distinguish between the meibomian glands and the conjunctiva structures in the acquired OCT images. First, we found the air-tissue boundary and removed the image of the air region. Second, we flattened each A-line signal of the OCT images to obtain the same relative height of the surface image. Finally, the image of the conjunctiva layer was removed with the same thickness for each OCT image and only the gland images were reconstructed. This image processing was carried out using a vision program developed based on MATLAB 7.1 (MathWorks, Natick, MA, USA).
The tomogram (B-scan, 5 mm) of the meibomian glands of subject A by FD-OCT is shown in
A: The spherical acini attached to a stalk were clearly visible in the 3D image; B: The 3D meibomian gland image was consistent with the meibomian glands in the rectangle in the infrared images.
A: There were extensive networks of branching structures between parallel groups of meibomian glands; B: We could not find the networks in the infrared images of meibomian gland.
A: The 3D images showed a few meibomian glands with a normal grape-like pattern. We could not find definite acini attached to the central ducts. B: These findings are consistent with the noncontact infrared meibography, which showed nearly complete loss of the meibomian glands.
A: Although the extent of the meibomian gland loss seemed minimal, we saw definite atrophic changes in the acini attached to the central ducts. The contours of the acini were ill-defined and the acini small; B: These findings are consistent with noncontact meibography showing ill-defined acini.
A: We found few meibomian glands with normal morphology, but observed spindle-shaped or globular structures instead; B: The 3D meibomian gland image was consistent with the meibomian glands in the rectangle in the infrared image.
Patient | Sex | Age | Grading |
Findings | Type | |||
Drop -out | Shortening | Distortion | Dilation | |||||
C | Male | 67 | 3 | o | o | Obstructive | ||
D | Female | 68 | 0–1 | o | o | Seborheic | ||
E | Male | 31 | 3 | o | o | o | Obstructive | |
F | Male | 50 | 2 | o | o | Seborheic | ||
G | Female | 70 | 1 | o | o | o | Seborheic | |
H | Male | 56 | 3 | o | o | o | o | Obstructive |
I | Female | 53 | 1 | o | o | o | Seborheic |
Meiboscore system by Arita et al
The thickness of the removed layer was approximately 345 µm (50 pixel height for
This is the first study in which
Recently, Bizheva
Recently, Matsumoto
We were able to successfully create a 3D image of the meibomian gland using our FD-OCT in this study because of its two major features. First, it is a high speed OCT, which was possible because it is a Fourier-Domain type OCT and uses a high-speed wavelength swept laser. Unlike corneas, everted eyelids can cause motion artifacts due to eyelid movement by the subject or by the finger of the examiner. The A-scan speed of the OCT used in this study was 52 kHz, which is 26 times faster than that of the commercial AS-OCT. It took only 4.2 seconds to scan a 5 × 2 mm area of the upper lid. It was impossible to take 200 tomograms for a 3D image without motion artifacts using the previous AS-OCT, which would require over 166.4 seconds. Second, it uses a light source with a central wavelength of 1310 nm. Because infrared light of 1310 nm can penetrate deeply into tissues, good images of the meibomian glands under the conjunctiva were obtained.
However, some problems remain. First, although the FD-OCT is 26 times faster than the previous AS-OCT, it is still slow in scanning the everted eyelid. During the 4.2 seconds necessary for scanning, minimizing the motion arising from the subject or the examiner is difficult. During examination, everted upper lids have forward-backward, up-down, and left-right movements. At post-processing, forward-backward movement could be partially corrected by the alignment of conjunctival surfaces in the tomograms (B-scans) with each other. However, up-down and left-right movement could not be corrected because there was no reference available. Some commercial OCT instruments have the capability to stabilize during ocular imaging. The OCT system incorporates an aiming target, which the subjects can look at during imaging. However, in 3D OCT meibography, the subject cannot look at the target because of everted upper lids. Furthermore, commercial OCTs have an auto-tracking function, which stabilizes images. In contrast, our system lacks auto-tracking. Addition of a tracking function would greatly reduce motion artifacts.
Second, the resolution of the 3D images remains low. Although the 3D images show more detailed structures than infrared meibography, the contrast is insufficient to identify the gland boundary. To enhance the image resolution, the optimal wavelength with minimum scattering at the conjunctiva should be determined; additionally, attempt other optical methods should be attempted.
Third, we must determine the potential clinical applications of 3D images of the meibomian glands. This OCT system shows the 3D morphology of the meibomian glands in detail. It might be possible to determine the type of MGD (obstructive or hyposecretory) using 3D images. Using noncontact meibography, we can devise a more detailed grading system for MGD than the meiboscore grade.
Similar to previous reports, the obstructive MGD group had a significantly higher average meiboscore than the control group, and various meibomian gland changes, such as dropout, shortening, distortion, and gland dilation, were seen in all patients in the obstructive MGD group.
Using the 3D images, we need to develop indices (
Fourth, the imaging field (5 × 2 mm) of this system is small compared to the size of the upper eyelids. Meibomian glands loss is often uneven across the eyelid. The scanned area might not be sufficiently large to view lost glands. The 3D morphology of the meibomian glands in the imaging field is not always representative of all of the meibomian glands in the eyelid. In patients with moderate MGD, the glands in the imaging field might even look relatively healthy. Therefore, the positioning of the field appears critical and might produce misleading results. Thus this technique would be useful for MGD only if used in conjunction with noncontact infrared meibography.
We needed two examiners to perform 3D OCT meibography. One examiner held the subject's everted upper lid and the other scans the palpebral conjunctiva using the OCT. The image quality of 3D OCT meibography varies depending on the examiner who held the upper lids (inter-examiner repeatability). The larger the involuntary hand tremor of the examiner, the greater the motion artifact. Even for the same examiner, the OCT meibography varied depending on the position of the imaging field or hand tremor (intra-examiner repeatability).
In this study, we performed 3D OCT meibography of the upper lids. Imaging of the lower eyelids is also possible. However, since the subjects blink their eyes frequently, the motion artifact with the lower eyelids is much greater than that of the upper lids in our experience. In addition, the meibomian glands of the lower lids are shorter than those of the upper eyelids. Therefore, it was sometimes difficult to find the meibomian glands in the small scan field.
In subject B, we found extensive networks of branching structures between parallel groups of meibomian glands (
A tomogram of the meibomian glands using OCT (
We successfully produced 3D images of human meibomian glands by reconstructing tomograms of these glands with high speed Fourier-Domain OCT developed in our laboratory. We believe that this imaging technique can be applied to research on meibomian glands.
(WMV)
(MPG)