Jui-Kai Wang

Multimodal Segmentation of Optic Disc and Cup From SD-OCT and Color Fundus Photographs Using a Machine-Learning Graph-Based Approach

In this work, a multimodal approach is proposed to use the complementary information from fundus photographs and spectral domain optical coherence tomography (SD-OCT) volumes in order to segment the optic disc and cup boundaries. The problem is formulated as an optimization problem where the optimal solution is obtained using a machine-learning theoretical graph-based method. In particular, first the fundus photograph is registered to the 2D projection of the SD-OCT volume. Three in-region cost functions are designed using a random forest classifier corresponding to three regions of cup, rim, and background. Next, the volumes are resampled to create radial scans in which the Bruchs Membrane Opening (BMO) endpoints are easier to detect. Similar to in-region cost function design, the disc-boundary cost function is designed using a random forest classifier for which the features are created by applying the Haar Stationary Wavelet Transform (SWT) to the radial projection image. A multisurface graph-based approach utilizes the in-region and disc-boundary cost images to segment the boundaries of optic disc and cup under feasibility constraints. The approach is evaluated on 25 multimodal image pairs from 25 subjects in a leave-one-out fashion (by subject). The performances of the graph-theoretic approach using three sets of cost functions are compared: 1) using unimodal (OCT only) in-region costs, 2) using multimodal in-region costs, and 3) using multimodal in-region and disc-boundary costs. Results show that the multimodal approaches outperform the unimodal approach in segmenting the optic disc and cup.

Retinal ganglion cell layer thinning within one month of presentation for optic neuritis

Background: Spectral domain optical coherence tomography (SD-OCT) reveals retinal ganglion cell layer plus inner plexiform layer (GCL+IPL) and peripapillary retinal nerve fiber layer (pRNFL) thinning in chronic optic nerve injury. At presentation, swelling of the pRNFL confounds evaluation of early axon loss. Objective: We studied whether the GCL+IPL thins before the pRNFL, the trajectory of GCL+IPL loss and relationship to vision. Methods: We prospectively evaluated 33 eyes (study) with new optic neuritis, using perimetry and SD-OCT with investigative three-dimensional layer segmentation and commercial two-dimensional segmentation to compute the GCL+IPL and pRNFL thickness. Results: At presentation, GCL+IPL thickness (82.4±8.8 µm) did not differ from unaffected fellow eyes (81.2±6.7 µm), via the three-dimensional method, while the two-dimensional method failed in 9% of study eyes. At 1–2 months, there was thinning of the pRNFL in 10% and of the GCL+IPL in 93% of study eyes. GCL+IPL reduction was greatest during the first 2 months. GCL+IPL thinning at 1–2 months correlated with GCL+IPL thinning at 6 months (r=0.84, P=0.01) and presentation visual acuity (r=0.48, P=0.006) and perimetric mean deviation (r=0.52, P=0.003). Conclusion: GGL+IPL is an early biomarker of structural injury in optic neuritis as thinning develops within 1–2 months of onset, prior to pRNFL thinning.

Retinal and Choroidal Folds in Papilledema.

PURPOSE To determine the frequency, patterns, associations, and biomechanical implications of retinal and choroidal folds in papilledema due to idiopathic intracranial hypertension (IIH). METHODS Retinal and choroidal folds were studied in patients enrolled in the IIH Treatment Trial using fundus photography (n = 165 study eyes) and spectral-domain optical coherence tomography (SD-OCT; n = 125). We examined the association between folds and peripapillary shape, retinal nerve fiber layer (RNFL) thickness, disc volume, Frisén grade, acuity, perimetric mean deviation, intraocular pressure, intracranial pressure, and refractive error. RESULTS We identified three types of folds in IIH patients with papilledema: peripapillary wrinkles (PPW), retinal folds (RF), and choroidal folds (CF). Frequency, with photos, was 26%, 19%, and 1%, respectively; SD-OCT frequency was 46%, 47%, and 10%. At least one type of fold was present in 41% of patients with photos and 73% with SD-OCT. Spectral-domain OCT was more sensitive. Structural parameters related to the severity of papilledema were associated with PPW and RF, whereas anterior deformation of the peripapillary RPE/basement membrane layer was associated with CF and RF. Folds were not associated with vision loss at baseline. CONCLUSIONS Folds in papilledema are biomechanical signs of stress/strain on the optic nerve head and load-bearing structures induced by intracranial hypertension. Folds are best imaged with SD-OCT. The patterns of retinal and choroidal folds are the products of a complex interplay between the degree of papilledema and anterior deformation of the load-bearing structures (sclera and possibly the lamina cribrosa), both modulated by structural geometry and material properties of the optic nerve head. (ClinicalTrials.gov number, NCT01003639.).

Causes and Prognosis of Visual Acuity Loss at the Time of Initial Presentation in Idiopathic Intracranial Hypertension.

PURPOSE To determine the etiology and prognosis of visual acuity loss in idiopathic intracranial hypertension (IIH) at presentation and to provide objective measures to predict visual outcome. METHODS A retrospective review of 660 patients with IIH (2009-2013) identified 31 patients (4.7%) with 48 eyes having best-corrected visual acuity (BCVA) of 20/25 or worse on initial presentation. Fundus photography, optical coherence tomography (OCT) of the optic disc and macula, and perimetry were used to determine the causes and prognosis of vision loss. Segmentation of the macula OCT was performed using the Iowa Reference Algorithm to determine the retinal ganglion cell-inner plexiform layer complex (GCL-IPL) thickness. RESULTS Outer retinal changes alone caused decreased BCVA at initial presentation in 22 eyes (46%): subretinal fluid in 16, chorioretinal folds in 5, and peripapillary choroidal neovascularization in 1. The vision loss was reversible except for some eyes with chorioretinal folds. Optic neuropathy alone caused decreased BCVA in 10 eyes (21%) and coexisting outer retinal changes and optic neuropathy caused decreased BCVA in 16 eyes (33%). A GCL-IPL thickness less than or equal to 70 μm at initial presentation or progressive thinning of greater than or equal to 10 μm within 2 to 3 weeks compared with baseline correlated with poor visual outcome. CONCLUSIONS Visual acuity loss in IIH can be caused by both outer retinal changes and optic neuropathy. Vision loss from outer retinal changes is mostly reversible. The outcome of patients with coexisting outer retinal changes and optic neuropathy or optic neuropathy alone depends on the degree of optic neuropathy, which can be predicted by the GCL-IPL thickness.

Quantitative evaluation of papilledema from stereoscopic color fundus photographs.

PURPOSE To derive a computerized measurement of optic disc volume from digital stereoscopic fundus photographs for the purpose of diagnosing and managing papilledema. METHODS Twenty-nine pairs of stereoscopic fundus photographs and optic nerve head (ONH) centered spectral domain optical coherence tomography (SD-OCT) scans were obtained at the same visit in 15 patients with papilledema. Some patients were imaged at multiple visits in order to assess their changes. Three-dimensional shape of the ONH was estimated from stereo fundus photographs using an automated multi-scale stereo correspondence algorithm. We assessed the correlation of the stereo volume measurements with the SD-OCT volume measurements quantitatively, in terms of volume of retinal surface elevation above a reference plane and also to expert grading of papilledema from digital fundus photographs using the Frisén grading scale. RESULTS The volumetric measurements of retinal surface elevation estimated from stereo fundus photographs and OCT scans were positively correlated (correlation coefficient r(2) = 0.60; P < 0.001) and were positively correlated with Frisén grade (Spearman correlation coefficient r = 0.59; P < 0.001). CONCLUSIONS Retinal surface elevation among papilledema patients obtained from stereo fundus photographs compares favorably with that from OCT scans and with expert grading of papilledema severity. Stereoscopic color imaging of the ONH combined with a method of automated shape reconstruction is a low-cost alternative to SD-OCT scans that has potential for a more cost-effective diagnosis and management of papilledema in a telemedical setting. An automated three-dimensional image analysis method was validated that quantifies the retinal surface topography with an imaging modality that has lacked prior objective assessment.

Semi-automated 2D Bruch's membrane shape analysis in papilledema using spectral-domain optical coherence tomography

Recent studies have shown that the Bruchs membrane (BM) and retinal pigment epithelium (RPE), visualized on spectral-domain optical coherence tomography (SD-OCT), is deformed anteriorly towards the vitreous in patients with intracranial hypertension and papilledema. The BM/RPE shape has been quantified using a statistical-shape-model approach; however, to date, the approach has involved the tedious and time-consuming manual placement of landmarks and correspondingly, only the shape (and shape changes) of a limited number of patients has been studied. In this work, we first present a semi-automated approach for the extraction of 20 landmarks along the BM from an optic-nerve-head (ONH) centered OCT slice from each patient. In the approach, after the manual placement of the two Bruchs membrane opening (BMO) points, the remaining 18 landmarks are automatically determined using a graph-based segmentation approach. We apply the approach to the OCT scans of 116 patients (at baseline) enrolled in the Idiopathic Intracranial Hypertension Treatment Trial and generate a statistical shape model using principal components analysis. Using the resulting shape model, the coefficient (shape measure) corresponding to the second principal component (eigenvector) for each set of landmarks indicates the degree of the BM/RPE is oriented away from the vitreous. Using a subset of 20 patients, we compare the shape measure computed using this semi-automated approach with the resulting shape measure when (1) all landmarks are specified manually (Experiment I); and (2) a different expert specifies the two BMO points (Experiment II). In each case, a correlation coefficient ≥ 0.99 is obtained.

Retinal Ganglion Cell Layer Thinning Within One Month of Presentation for Non-Arteritic Anterior Ischemic Optic Neuropathy

Purpose Optical coherence tomography reveals retinal ganglion cell layer (GCL) and retinal nerve fiber layer (RNFL) thinning in chronic optic nerve injury. With acute optic nerve injury, as in acute nonarteritic anterior ischemic optic neuropathy (NAION), swelling obscures early demonstration of RNFL thinning, which might be used to evaluate therapies. We hypothesized that measurement of GCL plus inner plexiform layer (IPL) thickness and trajectory of thinning would show it is an earlier and more accurate biomarker of early permanent neuronal injury. Methods We prospectively studied 29 acute NAION eyes with standard automated perimetry and spectral domain (SD) optical coherence tomography for 6 months. We used a three-dimensional layer segmentation (method 1) and a commercial proprietary (method 2), to compute the combined thickness of macular GCL+IPL and method 2 to compute peripapillary RNFL thickness. Results At presentation, the mean GCL+IPL thickness (78.7 μm ± 8.9) for NAION eyes, did not differ from unaffected fellow eyes (83 μm ± 6.4), using method 1 while method 2 (66.8 μm ± 18.7) failed in 34% of NAION eyes. At 1 to 2 months, 12% had RNFL loss compared to baseline, while 68% of NAION eyes had GCL+IPL thinning. The ganglion cell layer plus inner plexiform layer reduction was greatest at 1 to 2 months (19.6 μm ± 12.6) and was minimally worse after month 3. Ganglion cell layer plus inner plexiform layer thinning showed moderate to strong significant correlation with the visual acuity and mean deviation at each exam time. The retinal nerve fiber layer was not thinned until month 3. Conclusions Ganglion cell layer plus inner plexiform layer is acutely unaffected and provides a reliable measure of retinal neuronal structure using three-dimensional segmentation. Thinning develops within 1 to 2 months of onset, which is prior to RNFL swelling resolution. This suggests GCL+IPL measurement is better than the RNFL thickness to use as biomarker of early structural loss in NAION.

Determining degree of optic nerve edema from color fundus photography

Swelling of the optic nerve head (ONH) is subjectively assessed by clinicians using the Frisén scale. It is believed that a direct measurement of the ONH volume would serve as a better representation of the swelling. However, a direct measurement requires optic nerve imaging with spectral domain optical coherence tomography (SD-OCT) and 3D segmentation of the resulting images, which is not always available during clinical evaluation. Furthermore, telemedical imaging of the eye at remote locations is more feasible with non-mydriatic fundus cameras which are less costly than OCT imagers. Therefore, there is a critical need to develop a more quantitative analysis of optic nerve swelling on a continuous scale, similar to SD-OCT. Here, we select features from more commonly available 2D fundus images and use them to predict ONH volume. Twenty-six features were extracted from each of 48 color fundus images. The features include attributes of the blood vessels, optic nerve head, and peripapillary retina areas. These features were used in a regression analysis to predict ONH volume, as computed by a segmentation of the SD-OCT image. The results of the regression analysis yielded a mean square error of 2.43 mm3 and a correlation coefficient between computed and predicted volumes of R = 0:771, which suggests that ONH volume may be predicted from fundus features alone.

Automated surface segmentation of internal limiting membrane in spectral-domain optical coherence tomography volumes with a deep cup using a 3-D range expansion approach

Spectral-domain optical coherence tomography (SD-OCT) is used clinically for the diagnosis and management of glaucoma, with the internal limiting membrane (ILM) being one important structure of interest as it reflects the upper bounding surface of the optic cup. In cases of severe glaucoma, the boundary of the optic cup can become very steep and difficult to segment. Because of this difficulty, current automated surface segmentation algorithms will frequently fail in cases of deep cupping and provide a cup that is too shallow. In this work, we propose a 3-D graph-theoretic segmentation method based on the range expansion algorithm for accurate ILM segmentation in SD-OCT volumes of subjects with severe glaucoma. The performance of this approach was validated on 10 optic nerve head (ONH) centered OCT volumes obtained from 10 glaucomatous subjects. We computed the unsigned mean surface positioning (UMSP) error and unsigned average symmetric surface distance (UASSD) error with respect to manual tracings (ground truth) from one independent expert observer. Comparing to the widely used graph search method, our proposed method achieved a significant improvement in the UMSP error which was reduced from 7.69±3.23 μm to 6.35±1.35 μm (p <; 0.04), and in the UASSD error which was reduced from 3.86±0.54 μm to 3.40±0.17μm (p <; 0.07). It has high potential for future extension to multiple surface segmentation and clinical applicability.

The Pattern of Visual Fixation Eccentricity and Instability in Optic Neuropathy and Its Spatial Relationship to Retinal Ganglion Cell Layer Thickness

Purpose The purpose of this study was to assess whether clinically useful measures of fixation instability and eccentricity can be derived from retinal tracking data obtained during optical coherence tomography (OCT) in patients with optic neuropathy (ON) and to develop a method for relating fixation to the retinal ganglion cell complex (GCC) thickness. Methods Twenty-nine patients with ON underwent macular volume OCT with 30 seconds of confocal scanning laser ophthalmoscope (cSLO)-based eye tracking during fixation. Kernel density estimation quantified fixation instability and fixation eccentricity from the distribution of fixation points on the retina. Preferred ganglion cell layer loci (PGCL) and their relationship to the GCC thickness map were derived, accounting for radial displacement of retinal ganglion cell soma from their corresponding cones. Results Fixation instability was increased in ON eyes (0.21 deg2) compared with normal eyes (0.06982 deg2; P < 0.001), and fixation eccentricity was increased in ON eyes (0.48°) compared with normal eyes (0.24°; P = 0.03). Fixation instability and eccentricity each correlated moderately with logMAR acuity and were highly predictive of central visual field loss. Twenty-six of 35 ON eyes had PGCL skewed toward local maxima of the GCC thickness map. Patients with bilateral dense central scotomas had PGCL in homonymous retinal locations with respect to the fovea. Conclusions Fixation instability and eccentricity measures obtained during cSLO-OCT assess the function of perifoveal retinal elements and predict central visual field loss in patients with ON. A model relating fixation to the GCC thickness map offers a method to assess the structure–function relationship between fixation and areas of preserved GCC in patients with ON.