High frequency ultrasound biomicroscopy (UBM) images are used in clinical ophthalmology due to its ability to penetrate opaque tissues and create high resolution images of deeper intraocular structures. Because these inexpensive, high frequency (50 MHz) systems use single ultrasound elements, there is a limitation in visualizing small structures and anatomical landmarks, especially outside focal area, due to the lack of dynamic focusing. The wide and axially variant point spread function degrade image quality and obscure smaller structures. We created a fast, generative adversarial network (GAN) method to apply axially varying deconvolution for our 3D ultrasound biomicroscopy (3D-UBM) imaging system. Original images are enhanced using a computationally expensive axially varying deconvolution, giving paired original and enhanced images for GAN training. Supervised generative adversarial networks (pix2pix) were trained to generate enhanced images from originals. We obtained good performance metrics (SSIM = 0.85 and PSNR = 31.32 dB) in test images without any noticeable artifacts. GAN deconvolution runs at about 31 msec per frame on a standard graphics card, indicating that near real time enhancement is possible. With GAN enhancement, important ocular structures are made more visible.
We developed a methodology for 3D assessment of ciliary body of the eye, an important, but understudied tissue, using our new 3D ultrasound biomicroscopy (3D-UBM) imaging system. The ciliary body produces aqueous humor, which if not drained properly, can lead to increased intraocular pressure and glaucoma, a leading cause of blindness. Most medications and some surgical procedures for glaucoma target the ciliary body. Ciliary body is also responsible for focusing-accommodation by muscle contraction and relaxation. UBM is the only imaging modality which can be used to visualize structures behind the opaque iris, such as ciliary body. Our 3D-UBM acquires several hundred high resolutions (50 MHz) 2D-UBM images and creates a 3D volume, enabling heretofore unavailable en face visualizations and quantifications. In this study, we calculated unique 3D biometrics from automated segmentation using deep learning (UNet). Our results show accuracy of 0.93 ± 0.01, sensitivity of 0.79 ± 0.07 and dice score of 0.72 ± 0.07 on deep learning segmentation of ciliary muscle. For an eye, volume of ciliary body was 67.87 mm3, single ciliary process volumes were 0.234 ± 0.093 mm3 with surface areas adjacent to aqueous humor of 3.02 ± 1.07 mm2. Automated and manual measurements of ciliary muscle volume and cross-sectional area are compared which show overestimation in volume measurement but higher agreeability in cross-sectional area measurements.
KEYWORDS: Eye, 3D image processing, Ultrasonography, Image segmentation, Computer programming, 3D metrology, Stereoscopy, Iris, In vivo imaging, Optical coherence tomography
We created a new high resolution (50-MHz) three-dimensional ultrasound biomicroscopy (3D-UBM) imaging system and applied it to the measurement of iridoconeal angle, an important biomarker for glaucoma patients. Glaucoma, a leading cause of blindness, often results from poor drainage of the fluid from the eye through structures located at the iridiocorneal angle. Measurement of the angle has important implications for predicting the course of the disease and determining treatment strategies. An angle measured at a particular location with conventional 2D-UBM can be biased due to tilt in the hand-held probe. We created a 3D-UBM system by automatically scanning a 2D UBM with a precision translating stage. Using 3D-UBM, we typically acqure several hundred 2D images to create a high-resolution volume of the anterior chamber of the eye. Image pre-processing included intensity based frame-to-frame alignment to reduce effects of eye motion, 3D noise reduction, and multi-planar reformatting to create rotational views along the optic-axis with the pupil at the center, thereby giving views suitable for measurement of the iridiocorneal angle. Anterior chambers were segmented using a semantic-segmentation convolutional neural network, which gave folded “leave-one-eye-out” accuracy of 98.04%±0.01%, sensitivity of 90.97%±0.02%, specificity of 98.91%±0.01%, and Dice coefficient of 0.91±0.04. Using segmentations, iridiocorneal angles were automatically estimated using a modification of the semi-automated trabecular- iris-angle method (TIA) for each of ∼360 rotational views. Automated measurements were compared to those made by four ophthalmologist readers in eight images from two eyes. In these images, an insignificant difference (p = 0.996) was shown between readers and automated results.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.