Involuntary eye motion correction in retinal optical coherence tomography: Hardware or software solution?

OCT Segmentation Errors with Bruch's Membrane Opening-Minimum Rim Width as Compared with Retinal Nerve Fiber Layer Thickness

OBJECTIVE

To compare the magnitude and location of automated segmentation errors of the Bruch's membrane opening-minimum rim width (BMO-MRW) and retinal nerve fiber layer thickness (RNFLT).

DESIGN

Cross-sectional study.

PARTICIPANTS

We included 162 glaucoma suspect or open-angle glaucoma eyes from 162 participants.

METHODS

We used spectral-domain optic coherence tomography (Spectralis 870 nm, Heidelberg Engineering) to image the optic nerve with 24 radial optic nerve head B-scans and a 12-degree peripapillary circle scan, and exported the native "automated segmentation only" results for BMO-MRW and RNFLT. We also exported the results after "manual refinement" of the measurements.

MAIN OUTCOME MEASURES

We calculated the absolute and proportional error globally and within the 12 30-degree sectors of the optic disc. We determined whether the glaucoma classifications were different between BMO-MRW and RNFLT as a result of manual and automatic segmentation.

RESULTS

The absolute error mean was larger for BMO-MRW than for RNFLT (10.8 μm vs. 3.58 μm, P < 0.001). However, the proportional errors were similar (4.3% vs. 4.4%, P = 0.47). In a multivariable regression model, errors in BMO-MRW were not significantly associated with age, location, magnitude, or severity of glaucoma loss (all P ≥ 0.05). However, larger RNFLT errors were associated with the superior and inferior sector location, thicker nerve fiber layer, and worse visual field (all P < 0.05). Errors in BMO-MRW and RNFLT were not likely to occur in the same sector location (R2 = 0.001; P = 0.15). With manual refinement, the glaucoma classification changed in 7.8% and 6.2% of eyes with BMO-MRW and RNFLT, respectively.

CONCLUSIONS

Both BMO-MRW and RNFLT measurements included segmentation errors, which did not seem to have a common location, and may result in differences in glaucoma classification.

FINANCIAL DISCLOSURE(S)

Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

Spiral scanning improves subject fixation in widefield retinal imaging

Point scanning retinal imaging modalities, including confocal scanning light ophthalmoscopy (cSLO) and optical coherence tomography, suffer from fixational motion artifacts. Fixation targets, though effective at reducing eye motion, are infeasible in some applications (e.g., handheld devices) due to their bulk and complexity. Here, we report on a cSLO device that scans the retina in a spiral pattern under pseudo-visible illumination, thus collecting image data while simultaneously projecting, into the subject's vision, the image of a bullseye, which acts as a virtual fixation target. An imaging study of 14 young adult volunteers was conducted to compare the fixational performance of this technique to that of raster scanning, with and without a discrete inline fixation target. Image registration was used to quantify subject eye motion; a strip-wise registration method was used for raster scans, and a novel, to the best of our knowledge, ring-based method was used for spiral scans. Results indicate a statistically significant reduction in eye motion by the use of spiral scanning as compared to raster scanning without a fixation target.

Simultaneous alignment and surface regression using hybrid 2D-3D networks for 3D coherent layer segmentation of retinal OCT images with full and sparse annotations

Layer segmentation is important to quantitative analysis of retinal optical coherence tomography (OCT). Recently, deep learning based methods have been developed to automate this task and yield remarkable performance. However, due to the large spatial gap and potential mismatch between the B-scans of an OCT volume, all of them were based on 2D segmentation of individual B-scans, which may lose the continuity and diagnostic information of the retinal layers in 3D space. Besides, most of these methods required dense annotation of the OCT volumes, which is labor-intensive and expertise-demanding. This work presents a novel framework based on hybrid 2D-3D convolutional neural networks (CNNs) to obtain continuous 3D retinal layer surfaces from OCT volumes, which works well with both full and sparse annotations. The 2D features of individual B-scans are extracted by an encoder consisting of 2D convolutions. These 2D features are then used to produce the alignment displacement vectors and layer segmentation by two 3D decoders coupled via a spatial transformer module. Two losses are proposed to utilize the retinal layers' natural property of being smooth for B-scan alignment and layer segmentation, respectively, and are the key to the semi-supervised learning with sparse annotation. The entire framework is trained end-to-end. To the best of our knowledge, this is the first work that attempts 3D retinal layer segmentation in volumetric OCT images based on CNNs. Experiments on a synthetic dataset and three public clinical datasets show that our framework can effectively align the B-scans for potential motion correction, and achieves superior performance to state-of-the-art 2D deep learning methods in terms of both layer segmentation accuracy and cross-B-scan 3D continuity in both fully and semi-supervised settings, thus offering more clinical values than previous works.

Shannon entropy as a reliable score to diagnose human fibroelastic degenerative mitral chords: A micro-ct ex-vivo study

This paper is aimed at identifying by means of micro-CT the microstructural differences between normal and degenerative mitral marginal chordae tendineae. The control group is composed of 21 normal chords excised from 14 normal mitral valves from heart transplant recipients. The experimental group comprises 22 degenerative fibroelastic chords obtained at surgery from 11 pathological valves after mitral repair or replacement. In the control group the superficial endothelial cells and spongiosa layer remained intact, covering the wavy core collagen. In contrast, in the experimental group the collagen fibers were arranged as straightened thick bundles in a parallel configuration. 100 cross-sections were examined by micro-CT from each chord. Each image was randomized through the K-means machine learning algorithm and then, the global and local Shannon entropies were obtained. The optimum number of clusters, K, was estimated to maximize the differences between normal and degenerative chords in global and local Shannon entropy; the p-value after a nested ANOVA test was chosen as the parameter to be minimized. Optimum results were obtained with global Shannon entropy and 2≤K≤7, providing p < 0.01; for K=3, p = 2.86·10^-3. These findings open the door to novel perioperative diagnostic methods in order to avoid or reduce postoperative mitral valve regurgitation recurrences.

Extending field-of-view of retinal imaging by optical coherence tomography using convolutional Lissajous and slow scan patterns

Optical coherence tomography (OCT) is a high-speed non-invasive cross-sectional imaging technique. Although its imaging speed is high, three-dimensional high-spatial-sampling-density imaging of in vivo tissues with a wide field-of-view (FOV) is challenging. We employed convolved Lissajous and slow circular scanning patterns to extend the FOV of retinal OCT imaging with a 1-µm, 100-kHz-sweep-rate swept-source OCT prototype system. Displacements of sampling points due to eye movements are corrected by post-processing based on a Lissajous scan. Wide FOV three-dimensional retinal imaging with high sampling density and motion correction is achieved. Three-dimensional structures obtained using repeated imaging sessions of a healthy volunteer and two patients showed good agreement. The demonstrated technique will extend the FOV of simple point-scanning OCT, such as commercial ophthalmic OCT devices, without sacrificing sampling density.

Fourier spatial transform-based method of suppressing motion noises in OCTA

A large amount of lateral noise will be generated in blood flow imaging with optical coherence tomography angiography (OCTA) due to the presence of muscle shaking, heartbeat, and respiration, resulting in the deterioration of images. In this paper, to the best of our knowledge, for the first time, the spatial frequency information of motion noise in the blood flow signal region is used to remove the motion noise and false connections in the blood flow signal region. The effectiveness of the proposed adaptive denoising algorithm is verified by the imaging of finger blood flow. It is found that OCTA with different projection methods has improved signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) after applying our algorithm. It is also found that the visual effect of the original blood flow image based on standard deviation projection is better, but mean projection is the most sensitive to the algorithm, and the average SNR and CNR are improved by 5.7 dB and 8.9 dB, respectively.

Characterization and Analysis of Retinal Axial Motion at High Spatiotemporal Resolution and Its Implication for Real-Time Correction in Human Retinal Imaging

High-resolution ophthalmic imaging devices including spectral-domain and full-field optical coherence tomography (SDOCT and FFOCT) are adversely affected by the presence of continuous involuntary retinal axial motion. Here, we thoroughly quantify and characterize retinal axial motion with both high temporal resolution (200,000 A-scans/s) and high axial resolution (4.5 μm), recorded over a typical data acquisition duration of 3 s with an SDOCT device over 14 subjects. We demonstrate that although breath-holding can help decrease large-and-slow drifts, it increases small-and-fast fluctuations, which is not ideal when motion compensation is desired. Finally, by simulating the action of an axial motion stabilization control loop, we show that a loop rate of 1.2 kHz is ideal to achieve 100% robust clinical in-vivo retinal imaging.

LEARNING TO CORRECT AXIAL MOTION IN OCT FOR 3D RETINAL IMAGING

Optical Coherence Tomography (OCT) is a powerful technique for non-invasive 3D imaging of biological tissues at high resolution that has revolutionized retinal imaging. A major challenge in OCT imaging is the motion artifacts introduced by involuntary eye movements. In this paper, we propose a convolutional neural network that learns to correct axial motion in OCT based on a single volumetric scan. The proposed method is able to correct large motion, while preserving the overall curvature of the retina. The experimental results show significant improvements in visual quality as well as overall error compared to the conventional methods in both normal and disease cases.

Retinal cross-section motion correction in three-dimensional retinal optical coherence tomography

Motion correction is an important issue in ophthalmic optical coherence tomography (OCT), and can improve the ability of data sets to reflect the physiological structures of tissues and make visualization and subsequent analysis easier. In this study, we present a novel method to correct the cross-sectional motion artifacts in retinal OCT volumes. Motion along the x-direction (fast-scan direction) is corrected through the normalized cross-correlation algorithm, while axial motion compensation is performed using the polynomial fitting method on the inner segment/outer segment (IS/OS) layer segmented by the shortest path faster algorithm (SPFA). The results of volunteers with central serous chorioretinopathy demonstrate that the proposed method effectively corrects motion artifacts in OCT volumes and may have potential application value in the evaluation of ophthalmic diseases such as diabetic retinopathy, glaucoma and age-related macular degeneration.

Optical Coherence Tomography Artifacts Are Associated With Adaptive Optics Scanning Light Ophthalmoscopy Success in Achromatopsia

Purpose

To determine whether artifacts in optical coherence tomography (OCT) images are associated with the success or failure of adaptive optics scanning light ophthalmoscopy (AOSLO) imaging in subjects with achromatopsia (ACHM).

Methods

Previously acquired OCT and non-confocal, split-detector AOSLO images from one eye of 66 subjects with genetically confirmed achromatopsia (15 CNGA3 and 51 CNGB3) were reviewed along with best-corrected visual acuity (BCVA) and axial length. OCT artifacts in interpolated vertical volumes from CIRRUS macular cubes were divided into four categories: (1) none or minimal, (2) clear and low frequency, (3) low amplitude and high frequency, and (4) high amplitude and high frequency. Each vertical volume was assessed once by two observers. AOSLO success was defined as sufficient image quality in split-detector images at the fovea to assess cone quantity.

Results

There was excellent agreement between the two observers for assessing OCT artifact severity category (weighted kappa = 0.88). Overall, AOSLO success was 47%. For subjects with OCT artifact severity category 1, AOSLO success was 65%; for category 2, 47%; for category 3, 11%; and for category 4, 0%. There was a significant association between OCT artifact severity category and AOSLO success (P = 0.0002). Neither BCVA nor axial length was associated with AOSLO success (P = 0.07 and P = 0.75, respectively).

Conclusions

Artifacts in OCT volumes are associated with AOSLO success in ACHM. Subjects with less severe OCT artifacts are more likely to be good candidates for AOSLO imaging, whereas AOSLO was successful in only 7% of subjects with category 3 or 4 OCT artifacts. These results may be useful in guiding patient selection for AOSLO imaging.

Translational Relevance

Using OCT to prescreen patients could be a valuable tool for clinical trials that utilize AOSLO to reduce costs and decrease patient testing burden.

Accurately motion-corrected Lissajous OCT with multi-type image registration

Passive motion correction methods for optical coherence tomography (OCT) use image registration to estimate eye movements. To improve motion correction, a multi-image cross-correlation that employs spatial features in different image types is introduced. Lateral motion correction using en face OCT and OCT-A projections on Lissajous-scanned OCT data is applied. Motion correction using OCT-A projection of whole depth and OCT amplitude, OCT logarithmic intensity, and OCT maximum intensity projections were evaluated in retinal imaging with 76 patients. The proposed method was compared with motion correction using OCT-A projection of whole depth. The comparison shows improvements in the image quality of motion-corrected superficial OCT-A images and image registration.

Numerical method for axial motion artifact correction in retinal spectral-domain optical coherence tomography

A numerical method that compensates image distortions caused by random fluctuations of the distance to an object in spectral-domain optical coherence tomography (SD OCT) has been proposed and verified experimentally. The proposed method is based on the analysis of the phase shifts between adjacent scans that are caused by micrometer-scale displacements and the subsequent compensation for the displacements through phase-frequency correction in the spectral space. The efficiency of the method is demonstrated in model experiments with harmonic and random movements of a scattering object as well as during in vivo imaging of the retina of the human eye.

Matching Misaligned Spectralis OCTs to a Reference Scan in Pediatric Glaucoma with Poor Fixation and Nystagmus

Purpose

Poor fixation or nystagmus in children causes misalignment errors when measuring circumpapillary retinal nerve fiber layer (cpRNFL) thickness by simultaneous scanning laser ophthalmoscope imaging/optical coherence tomography (SLO/OCT). We investigated a method to assess cpRNFL from misaligned SLO/OCT scans.

Methods

Heidelberg Spectralis SLO/OCT scans from a single clinical examination were retrospectively analyzed when automated eye tracking was unreliable. Retinal layer thickness was measured at overlapping match locations between a reference and misaligned scans based on the position data from simultaneously acquired SLO images. Three layers were segmented: cpRNFL, internal limiting membrane to outer nuclear layer (ILM-ONL), and total retinal thickness (TR). Accuracy was defined as the difference in thickness between the reference and misaligned scans at their match locations after correction for scan angle.

Results

Thirty-five subjects, evaluated for glaucomatous nerve loss, met inclusion criteria. Group-averaged accuracy was −2.7, 1.4, and 0.3 µm for cpRNFL, ILM-ONL, and TR thickness, respectively. Across all layers, interobserver intraclass correlation coefficients ranged from 0.97 to 0.63 and the maximum Bland-Altman 95% limits of agreement were −21.6 to 20.7 µm. Variability was greatest for cpRNFL thickness and least for TR thickness. Increased variability was associated with lower signal-to-noise ratio but not with image-motion indices of shear, rotation, and scale.

Conclusions

Retinal layer thickness can be compared to a reference cpRNFL OCT scan when poor fixation and nystagmus causes misalignment errors. The analysis can be performed post hoc using multiple misaligned scans from standard SLO/OCT protocols.

Translational Relevance

Our method allows for assessment of cpRNFL in children who fail eye tracking.

In vivo noninvasive measurement of spatially resolved corneal elasticity in human eyes using Lamb wave optical coherence elastography

Current elastography techniques are limited in application to accurately assess spatially resolved corneal elasticity in vivo for human eyes. The air-puff optical coherence elastography (OCE) with an eye motion artifacts correction algorithm is developed to distinguish the in vivo cornea vibration from the eye motion and visualize the Lamb wave propagation clearly in healthy subjects. Based on the Lamb wave model, the phase velocity dispersion curve in the high-frequency is calculated to obtain spatially resolved corneal elasticity accurately with high repeatability. It is found that the corneal elasticity has regional variations and is correlated with intraocular pressure, which suggests that the method has the potential to provide noninvasive measurement of spatially resolved corneal elasticity in clinical practice.

Spatio-temporal filtering in laser Doppler holography for retinal blood flow imaging

Laser Doppler holography (LDH) is a full-field interferometric imaging technique recently applied in ophthalmology to measure blood flow, a parameter of high clinical interest. From the temporal fluctuations of digital holograms acquired at ultrafast frame rates, LDH reveals retinal and choroidal blood flow with a few milliseconds of temporal resolution. However, LDH experiences difficulties to detect slower blood flow as it requires to work with low Doppler frequency shifts which are corrupted by eye motion. We here demonstrate the use of a spatio-temporal decomposition adapted from Doppler ultrasound that provides a basis appropriate to the discrimination of blood flow from eye motion. A singular value decomposition (SVD) can be used as a simple, robust, and efficient way to separate the Doppler fluctuations of blood flow from those of strong spatial coherence such as eye motion. We show that the SVD outperforms the conventional Fourier based filter to reveal slower blood flow, and dramatically improves the ability of LDH to reveal vessels of smaller size or with a pathologically reduced blood flow.

Super-Resolution of Optical Coherence Tomography Images by Scale Mixture Models

In this paper, a new statistical model is proposed for the single image super-resolution of retinal Optical Coherence Tomography (OCT) images. OCT imaging relies on interfero-metry, which explains why OCT images suffer from a high level of noise. Moreover, data subsampling is carried out during the acquisition of OCT A-scans and B-scans. So, it is necessary to utilize effective super-resolution algorithms to reconstruct high-resolution clean OCT images. In this paper, a nonlocal sparse model-based Bayesian framework is proposed for OCT restoration. For this reason, by characterizing nonlocal patches with similar structures, known as a group, the sparse coefficients of each group of OCT images are modeled by the scale mixture models. In this base, the coefficient vector is decomposed into the point-wise product of a random vector and a positive scaling variable. Estimation of the sparse coefficients depends on the proposed distribution for the random vector and scaling variable where the Laplacian random vector and Generalized Extreme-Value (GEV) scale parameter (Laplacian+GEV model) show the best goodness of fit for each group of OCT images. Finally, a new OCT super-resolution method based on this new scale mixture model is introduced, where the maximum a posterior estimation of both sparse coefficients and scaling variables are calculated efficiently by applying an alternating minimization method. Our experimental results prove that the proposed OCT super-resolution method based on the Laplacian+GEV model outperforms other competing methods in terms of both subjective and objective visual qualities.

Assessment of corneal viscoelasticity using elastic wave optical coherence elastography

The corneal viscoelasticity have great clinical significance, such as the early diagnosis of keratoconus. In this work, an analysis method which utilized the elastic wave velocity, frequency and energy attenuation to assess the corneal viscoelasticity is presented. Using phase-resolved optical coherence tomography, the spatial-temporal displacement map is derived. The phase velocity dispersion curve and center frequency are obtained by transforming the displacement map into the wavenumber-frequency domain through the 2D fast Fourier transform (FFT). The shear modulus is calculated through Rayleigh wave equation using the phase velocity in the high frequency. The normalized energy distribution is plotted by transforming the displacement map into the spatial-frequency domain through the 1D FFT. The energy attenuation coefficient is derived by exponential fitting to calculate the viscous modulus. Different concentrations of tissue-mimicking phantoms and porcine corneas are imaged to validate this method, which demonstrates that the method has the capability to assess the corneal viscoelasticity.

In-vivo 3D corneal elasticity using air-coupled ultrasound optical coherence elastography

Corneal elasticity can resist elastic deformations under intraocular pressure to maintain normal corneal shape, which has a great influence on corneal refractive function. Elastography can measure tissue elasticity and provide a powerful tool for clinical diagnosis. Air-coupled ultrasound optical coherence elastography (OCE) has been used in the quantification of ex-vivo corneal elasticity. However, in-vivo imaging of the cornea remains a challenge. The 3D air-coupled ultrasound OCE with an axial motion artifacts correction algorithm was developed to distinguish the in-vivo cornea vibration from the axial eye motion in anesthetized rabbits and visualize the elastic wave propagation clearly. The elastic wave group velocity of in-vivo rabbit cornea was measured to be 5.96 ± 0.55 m/s, which agrees with other studies. The results show the potential of 3D air-coupled ultrasound OCE with an axial motion artifacts correction algorithm for quantitative in-vivo assessment of corneal elasticity.

Review on Retrospective Procedures to Correct Retinal Motion Artefacts in OCT Imaging

Motion artefacts from involuntary changes in eye fixation remain a major imaging issue in optical coherence tomography (OCT). This paper reviews the state-of-the-art of retrospective procedures to correct retinal motion and axial eye motion artefacts in OCT imaging. Following an overview of motion induced artefacts and correction strategies, a chronological survey of retrospective approaches since the introduction of OCT until the current days is presented. Pre-processing, registration, and validation techniques are described. The review finishes by discussing the limitations of the current techniques and the challenges to be tackled in future developments.

Three-dimensional optical coherence tomography image denoising through multi-input fully-convolutional networks

In recent years, there has been a growing interest in applying convolutional neural networks (CNNs) to low-level vision tasks such as denoising and super-resolution. Due to the coherent nature of the image formation process, the optical coherence tomography (OCT) images are inevitably affected by noise. This paper proposes a new method named the multi-input fully-convolutional networks (MIFCN) for denoising of OCT images. In contrast to recently proposed natural image denoising CNNs, the proposed architecture allows the exploitation of high degrees of correlation and complementary information among neighboring OCT images through pixel by pixel fusion of multiple FCNs. The parameters of the proposed multi-input architecture are learned by considering the consistency between the overall output and the contribution of each input image. The proposed MIFCN method is compared with the state-of-the-art denoising methods adopted on OCT images of normal and age-related macular degeneration eyes in a quantitative and qualitative manner.

Embedded deep learning in ophthalmology: making ophthalmic imaging smarter

Deep learning has recently gained high interest in ophthalmology due to its ability to detect clinically significant features for diagnosis and prognosis. Despite these significant advances, little is known about the ability of various deep learning systems to be embedded within ophthalmic imaging devices, allowing automated image acquisition. In this work, we will review the existing and future directions for 'active acquisition'-embedded deep learning, leading to as high-quality images with little intervention by the human operator. In clinical practice, the improved image quality should translate into more robust deep learning-based clinical diagnostics. Embedded deep learning will be enabled by the constantly improving hardware performance with low cost. We will briefly review possible computation methods in larger clinical systems. Briefly, they can be included in a three-layer framework composed of edge, fog, and cloud layers, the former being performed at a device level. Improved egde-layer performance via 'active acquisition' serves as an automatic data curation operator translating to better quality data in electronic health records, as well as on the cloud layer, for improved deep learning-based clinical data mining.