AxoNet 2.0: A Deep Learning-Based Tool for Morphometric Analysis of Retinal Ganglion Cell Axons.

Purpose: Assessment of glaucomatous damage in animal models is facilitated by rapid and accurate quantification of retinal ganglion cell (RGC) axonal loss and morphologic change. However, manual assessment is extremely time- and labor-intensive. Here, we developed AxoNet 2.0, an automated deep learning (DL) tool that (i) counts normal-appearing RGC axons and (ii) quantifies their morphometry from light micrographs. Methods: A DL algorithm was trained to segment the axoplasm and myelin sheath of normal-appearing axons using manually-annotated rat optic nerve (ON) cross-sectional micrographs. Performance was quantified by various metrics (e.g., soft-Dice coefficient between predicted and ground-truth segmentations). We also quantified axon counts, axon density, and axon size distributions between hypertensive and control eyes and compared to literature reports. Results: AxoNet 2.0 performed very well when compared to manual annotations of rat ON (R2 = 0.92 for automated vs. manual counts, soft-Dice coefficient = 0.81 ± 0.02, mean absolute percentage error in axonal morphometric outcomes < 15%). AxoNet 2.0 also showed promise for generalization, performing well on other animal models (R2 = 0.97 between automated versus manual counts for mice and 0.98 for non-human primates). As expected, the algorithm detected decreased in axon density in hypertensive rat eyes (P ≪ 0.001) with preferential loss of large axons (P < 0.001). Conclusions: AxoNet 2.0 provides a fast and nonsubjective tool to quantify both RGC axon counts and morphological features, thus assisting with assessing axonal damage in animal models of glaucomatous optic neuropathy. Translational Relevance: This deep learning approach will increase rigor of basic science studies designed to investigate RGC axon protection and regeneration.

AxoNet: A deep learning-based tool to count retinal ganglion cell axons.

In this work, we develop a robust, extensible tool to automatically and accurately count retinal ganglion cell axons in optic nerve (ON) tissue images from various animal models of glaucoma. We adapted deep learning to regress pixelwise axon count density estimates, which were then integrated over the image area to determine axon counts. The tool, termed AxoNet, was trained and evaluated using a dataset containing images of ON regions randomly selected from whole cross sections of both control and damaged rat ONs and manually annotated for axon count and location. This rat-trained network was then applied to a separate dataset of non-human primate (NHP) ON images. AxoNet was compared to two existing automated axon counting tools, AxonMaster and AxonJ, using both datasets. AxoNet outperformed the existing tools on both the rat and NHP ON datasets as judged by mean absolute error, R2 values when regressing automated vs. manual counts, and Bland-Altman analysis. AxoNet does not rely on hand-crafted image features for axon recognition and is robust to variations in the extent of ON tissue damage, image quality, and species of mammal. Therefore, AxoNet is not species-specific and can be extended to quantify additional ON characteristics in glaucoma and potentially other neurodegenerative diseases.

Optic nerve head blood flow response to reduced ocular perfusion pressure by alteration of either the blood pressure or intraocular pressure.

PURPOSE: To test the hypothesis that blood flow autoregulation in the optic nerve head has less reserve to maintain normal blood flow in the face of blood pressure-induced ocular perfusion pressure decrease than a similar magnitude intraocular pressure-induced ocular perfusion pressure decrease. MATERIALS AND METHODS: Twelve normal non-human primates were anesthetized by continuous intravenous infusion of pentobarbital. Optic nerve blood flow was monitored by laser speckle flowgraphy. In the first group of animals (n = 6), the experimental eye intraocular pressure was maintained at 10 mmHg using a saline reservoir connected to the anterior chamber. The blood pressure was gradually reduced by a slow injection of pentobarbital. In the second group (n = 6), the intraocular pressure was slowly increased from 10 mmHg to 50 mmHg by raising the reservoir. In both experimental groups, optic nerve head blood flow was measured continuously. The blood pressure and intraocular pressure were simultaneously recorded in all experiments. RESULTS: The optic nerve head blood flow showed significant difference between the two groups (p = 0.021, repeat measures analysis of variance). It declined significantly more in the blood pressure group compared to the intraocular pressure group when the ocular perfusion pressure was reduced to 35 mmHg (p < 0.045) and below. There was also a significant interaction between blood flow changes and the ocular perfusion pressure treatment (p = 0.004, adjusted Greenhouse & Geisser univariate test), indicating the gradually enlarged blood flow difference between the two groups was due to the ocular perfusion pressure decrease. CONCLUSIONS: The results show that optic nerve head blood flow is more susceptible to an ocular perfusion pressure decrease induced by lowering the blood pressure compared with that induced by increasing the intraocular pressure. This blood flow autoregulation capacity vulnerability to low blood pressure may provide experimental evidence related to the hemodynamic pathophysiology in glaucoma.

Relationship between orbital optic nerve axon counts and retinal nerve fiber layer thickness measured by spectral domain optical coherence tomography.

PURPOSE: We determined the relationship between total optic nerve axon counts and peripapillary retinal nerve fiber layer thickness (RNFLT) measured in vivo by spectral domain optical coherence tomography (SDOCT). METHODS: A total of 22 rhesus macaques had three or more baseline measurements in both eyes of peripapillary RNFLT made by SDOCT. Laser photocoagulation then was applied to the trabecular meshwork of one eye to induce chronic unilateral IOP elevation. SDOCT measurements of RNFLT continued approximately every two weeks until the predefined study endpoint was reached in each animal. At endpoint, animals were sacrificed and the optic nerve was sampled approximately 2 mm behind the globe to obtain thin sections for histologic processing and automated axon counting across 100% of the optic nerve cross-sectional area. RESULTS: At the final imaging session, the average loss of RNFLT was 20 ± 21%, ranging from essentially no loss to nearly 65% loss. Total optic nerve axon count in control eyes ranged from 812,478 to 1,280,474. The absolute number of optic nerve axons was related linearly to RNFLT (axon count = 12,336 × RNFLT((µm)) - 257,050, R(2) = 0.65, P < 0.0001), with a Pearson correlation coefficient of 0.81. There also was a strong linear relationship between relative optic nerve axon loss (glaucomatous-to-control eye) and relative RNFLT at the final imaging session, with a slope close to unity but a significantly negative intercept (relative axon loss((%)) = 1.05 × relative RNFLT loss((%)) - 14.4%, R(2) = 0.75, P < 0.0001). The negative intercept was robust to variations of fitted model because relative axon loss was -14% on average for all experimental glaucoma (EG) eyes within 6% (measurement noise) of zero relative loss. CONCLUSIONS: There is a strong linear relationship between total optic nerve axon count and RNFLT measured in vivo by SDOCT. However, substantial loss of optic nerve axons (∼10%-15%) exists before any loss of RNFLT manifests and this discrepancy persists systematically throughout a wide range of damage.

Anterior and posterior optic nerve head blood flow in nonhuman primate experimental glaucoma model measured by laser speckle imaging technique and microsphere method.

PURPOSE: To characterize optic nerve head (ONH) blood flow (BF) changes in nonhuman primate experimental glaucoma (EG) using laser speckle flowgraphy (LSFG) and the microsphere method and to evaluate the correlation between the two methods. METHODS: EG was induced in one eye each of 9 rhesus macaques by laser treatment to the trabecular meshwork. Prior to lasering and following onset of intraocular pressure (IOP) elevation, retinal never fiber layer thickness (RNFLT) and ONH BF were measured biweekly by spectral-domain optical coherence tomography and LSFG, respectively, until RNFLT loss was approximately 40% in the EG eye. Final BF was measured by LSFG and by the microsphere method in the anterior ONH (MS-BF(ANT)), posterior ONH (MS-BF(POST)), and peripapillary retina (MS-BF(PP)). RESULTS: Baseline RNFLT and LSFG-BF showed no difference between the two eyes (P = 0.69 and P = 0.43, respectively, paired t-test). Mean (± SD) IOP was 30 ± 6 mm Hg in EG eyes and 13 ± 2 mm Hg in control eyes (P < 0.001). EG eye RNFLT and LSFG-BF were reduced by 42 ± 16% (P < 0.0001) and 22 ± 13% (P = 0.003), respectively, at the final time point. EG eye MS-BF(ANT), MS-BF(POST), and MS-BF(PP) were reduced by 41 ± 17% (P < 0.001), 22 ± 34% (P = 0.06), and 30 ± 12% (P = 0.001), respectively, compared with the control eyes. Interocular ONH LSFG-BF differences significantly correlated to that measured by the microsphere method (R(2) = 0.87, P < 0.001). CONCLUSIONS: Chronic IOP elevation causes significant ONH BF decreases in the EG model. The high correlation between the BF reduction measured by LSFG and the microsphere method provides evidence that the LSFG is capable of assaying BF for a critical deep ONH region.

Structural and functional abnormalities of retinal ganglion cells measured in vivo at the onset of optic nerve head surface change in experimental glaucoma.

PURPOSE: To compare peripapillary retinal nerve fiber layer thickness (RNFLT), RNFL retardance, and retinal function at the onset of optic nerve head (ONH) surface topography change in experimental glaucoma (EG). METHODS: Thirty-three rhesus macaques had three or more weekly baseline measurements in both eyes of ONH surface topography, peripapillary RNFLT, RNFL retardance, and multifocal electroretinography (mfERG). Laser photocoagulation was then applied to the trabecular meshwork of one eye to induce chronic elevation of IOP and weekly recordings continued alternating between ONH surface topography and RNFLT during one week and RNFL retardance and mfERG the next week. Data were pooled for the group at the onset of ONH surface topography change in each EG eye, which was defined as the first date when either the mean position of the disc (MPD) fell below the 95% confidence limit of each eye's individual baseline range and/or when the topographic change analysis (TCA) map was subjectively judged as having demonstrated change, whichever came first. Analysis of variance with post hoc tests corrected for multiple comparisons were used to assess parameter changes. RESULTS: At onset of ONH surface topography change, there was no significant difference for RNFLT versus baseline or fellow control eyes. RNFL retardance and mfERG were significantly reduced in the recordings just prior (median of 9 days) to ONH onset (P < 0.01) and had progressed significantly (P < 0.001) an average of 17 days later (median of 7 days after ONH onset). RNFLT did not exhibit significant thinning until 15 days after onset of ONH surface topography change (P < 0.001). CONCLUSIONS: These results support the hypothesis that during the course of glaucomatous neurodegeneration, axonal cytoskeletal and retinal ganglion cell functional abnormalities exist before thinning of peripapillary RNFL axon bundles begins.

Deformation of the rodent optic nerve head and peripapillary structures during acute intraocular pressure elevation.

PURPOSE. To evaluate the effect of acutely elevated intraocular pressure (IOP) on retinal thickness and optic nerve head (ONH) structure in the rat eye by spectral domain-optical coherence tomography (SD-OCT). METHODS. Fourteen adult male Brown-Norway rats were studied under anesthesia (ketamine/xylazine/acepromazine, 55:5:1 mg/kg intramuscularly). Both eyes were imaged by SD-OCT on two baseline occasions several weeks before and again 2 and 4 weeks after the acute IOP imaging session. During the acute IOP session, SD-OCT imaging was performed 10 minutes after IOP was manometrically set at 15 mm Hg and then at 10, 30, and 60 minutes after IOP had been elevated to 50 mm Hg (n = 8) and again 10 and 30 minutes after IOP had been lowered back to 15 mm Hg (recovery). In two additional groups, IOP elevation was set to 70 mm Hg (n = 4) or 40 mm Hg (n = 2). Acute IOP results are reported for a pattern of 49 horizontal B-scans spanning a 20° square and follow-up results for peripapillary circular B-scans. Retinal and retinal nerve fiber layer (RNFL) thicknesses were measured with custom software by manual image segmentation. Friedman and Dunn's tests were used to assess acute and longer-term effects of acute IOP elevation. RESULTS. Acute IOP elevation to 50 mm Hg caused rapid (within seconds) deformation of the ONH and peripapillary structures, including posterior displacement of the ONH surface and outward bowing of peripapillary tissue; retinal thickness decreased progressively from 10 to 30 to 60 minutes by 16%, 18%, and 20% within the area of Bruch's membrane opening (BMO; P < 0.0001) by 8%, 9%, and 11% within the central 10° (excluding the BMO; P < 0.0001) but only by 1%, 2%, and 2.4% beyond the central 10° (P < 0.0001). Recovery was progressive and nearly complete by 30 minutes. Acute IOP elevation to 40 and 70 mm Hg produced similar structural changes, but 70 mm Hg also interfered with retinal blood flow. There were no changes in peripapillary retinal or RNFL thickness (P = 0.08 and P = 0.16, respectively) measured 2 and 4 weeks after acute elevation to 50 mm Hg. CONCLUSIONS. Acute IOP elevation in the rodent eye causes rapid, reversible posterior deformation of the ONH and thinning of the peripapillary retina, with only minimal retinal thinning beyond 5° of the ONH. No permanent changes in peripapillary retinal or RNFL thickness (for up to 1 month of follow-up) were caused by 60 minutes of IOP elevation to 50 mm Hg.

A comparison of optic nerve head morphology viewed by spectral domain optical coherence tomography and by serial histology.

PURPOSE: To compare serial optic nerve head (ONH) histology with interpolated B-scans generated from a three-dimensional (3-D) spectral domain (SD)-OCT ONH volume acquired in vivo from the same normal monkey eye. METHODS: A 15 degrees ONH SD-OCT volume was acquired in a normal monkey eye, with IOP manometrically controlled at 10 mm Hg. After perfusion fixation at 10 mm Hg, the ONH was trephined, the specimen embedded in a paraffin block, and serial sagittal sections cut at 4-mum intervals. The location of each histologic section was identified within the optic disc photograph by matching the position of the retinal vessels and of Bruch's membrane opening. By altering the angles of rotation and incidence, interpolated B-scans matching the location of the histologic sections were generated with custom software. Structures identified in the histologic sections were compared with signals identified in the matched B-scans. RESULTS: Close matches between histologic sections and interpolated B-scans were identified throughout the extent of the ONH. SD-OCT identified the neural canal opening as the termination of the Bruch's membrane-retinal pigment complex and border tissue as the innermost termination of the choroidal signal. The anterior lamina cribrosa and its continuity with the prelaminar glial columns were also detected by SD-OCT. CONCLUSIONS: Volumetric SD-OCT imaging of the ONH generates interpolated B-scans that accurately match serial histologic sections. SD-OCT captures the anterior laminar surface, which is likely to be a key structure in the detection of early ONH damage in ocular hypertension and glaucoma.

The effect of acute intraocular pressure elevation on peripapillary retinal thickness, retinal nerve fiber layer thickness, and retardance.

PURPOSE: To determine whether acutely elevated intraocular pressure (IOP) alters peripapillary retinal thickness, retinal nerve fiber layer thickness (RNFLT), or retardance. METHODS: Nine adult nonhuman primates were studied while under isoflurane anesthesia. Retinal and RNFLTs were measured by spectral domain optical coherence tomography 30 minutes after IOP was set to 10 mm Hg and 60 minutes after IOP was set to 45 mm Hg. RNFL retardance was measured by scanning laser polarimetry in 10-minute intervals for 30 minutes while IOP was 10 mm Hg, then for 60 minutes while IOP was 45 mm Hg, then for another 30 minutes after IOP was returned to 10 mm Hg. RESULTS: RNFLT measured 1120 microm from the ONH center decreased from 118.1 +/- 9.3 microm at an IOP of 10 mm Hg to 116.5 +/- 8.4 microm at 45 mm Hg, or by 1.4% +/- 1.8% (P < 0.0001). There was a significant interaction between IOP and eccentricity (P = 0.0006). Within 800 microm of the ONH center, the RNFL was 4.9% +/- 3.4% thinner 60 minutes after IOP elevation to 45 mm Hg (P < 0.001), but unchanged for larger eccentricities. The same pattern was observed for retinal thickness, with 1.1% +/- 0.8% thinning overall at 45 mm Hg (P < 0.0001), and a significant effect of eccentricity (P < 0.0001) whereby the retina was 4.8% +/- 1.2% thinner (P < 0.001) within 800 microm, but unchanged beyond that. Retardance increased by a maximum of 2.2% +/- 1.1% 60 minutes after IOP was increased to 45 mm Hg (P = 0.0031). CONCLUSIONS: The effects of acute IOP elevation on retinal thickness, RNFL thickness and retardance were minor, limited to the immediate ONH surround and unlikely to have meaningful clinical impact.

Relative course of retinal nerve fiber layer birefringence and thickness and retinal function changes after optic nerve transection.

PURPOSE: To test the hypothesis that alterations of RNFL birefringence precede changes in RNFL thickness in an experimental model of RGC injury and, secondarily, to determine the time course of RGC functional abnormalities relative to RNFL birefringence and thickness changes. METHODS: RNFL birefringence was measured by scanning laser polarimetry (GDx VCC; Carl Zeiss Meditec, Inc., Dublin, CA). RNFL thickness was measured by spectral domain optical coherence tomography (SD-OCT, Spectralis HRA+OCT; Heidelberg Engineering, GmbH, Heidelberg, Germany). Retinal function was assessed by three forms of electroretinography (ERG): slow-sequence multifocal (mf)ERG (VERIS; EDI, San Mateo, CA); pattern-reversal (P)ERG (Utas-E3000; LKC Technologies, Inc. Gaithersburg, MD); and photopic full-field flash (ff)ERG (Utas-E3000; LKC Technologies). All measurements were obtained in both eyes of four adult rhesus macaque monkeys (Macaca mulatta) during two baseline sessions, and again 1 week and 2 weeks after unilateral optic nerve transection (ONT). RESULTS: ONT was successfully completed in three subjects. RNFL birefringence declined by 15% 1 week after ONT (P = 0.043), whereas there was no significant change in RNFL thickness (+1%, P = 0.42). Two weeks after ONT, RNFL retardance had declined by 39% (P = 0.018), whereas RNFL thickness had declined by only 15% (P = 0.025). RGC functional abnormalities were present 1 week after ONT, including decreased amplitudes relative to baseline of the mfERG high-frequency components (-65%, P = 0.018), the PERG N95 component (-70%, P = 0.007), and the photopic negative response of the ffERG (-44%, P = 0.005). CONCLUSIONS: RNFL birefringence declined before and faster than RNFL thickness after ONT. RGC functional abnormalities were present 1 week after ONT, when RNFL thickness had not yet begun to change. RNFL birefringence changes after acute RGC injury are associated with RGC dysfunction. Together, they reflect RGC abnormalities that precede axonal caliber changes and loss.