Faculty Bibliography

Purpose : Current GPU memory limitations do not support the analysis of OCT scans at its original resolution, and previous techniques have downsampled the inputs considerably which resulted in a loss of detail. Here, we utilise a new memory management support framework that allows for the training of large deep learning networks and apply it to the detection of glaucoma in OCT scans at its original resolution. Methods : A total of 1110 SDOCT volumes (Cirrus, Zeiss, CA) were acquired from both eyes of 624 subjects (139 healthy and 485 glaucomatous patients (POAG)). A convolutional neural network (CNN) consisting of 8 3D-convolutional layers with a total of 600K parameters and was trained using a cross-entropy loss to differentiate between the healthy and glaucomatous scans. To avoid GPU memory constraints, the network was trained using a large model support library that automatically adds swap-in and swap-out nodes for transferring tensors from GPUs to the host and vice versa. This allowed for the OCT scans to be analysed at the original resolution of 200x200x1024. The performance of the network was gauged by computing the area under the receiver operating characteristic (AUC) curve. The performance of this network was also compared to a previously proposed network that ingested downsampled OCT scans (50x50x128), consisted of 5 3D-convolutional layers and had a total of 222K parameters; and a machine-learning technique (random forests) that relied on segmented features (peripapillary nerve fibre thicknesses). Class activation maps (CAM) were also generated for each of these networks to provide a qualitative view of the regions that the network deemed as important and relevant to the task. Results : The AUCs computed on the test set for the networks that analysed the volumes at the original and downsampled resolutions was found to be 0.92 and 0.91, respectively. The CAMs obtained using the high resolution images show more detail in comparison to the downsampled volume. The random forest technique showed an AUC of 0.85. Conclusions : The performance of the two networks was comparable for glaucoma detection but showed a vast improvement over the random forest that relied on segmented features. The ability to retain detail (as shown in the CAM) will likely allow for improvements in other tasks, such as spatial correspondences between visual field test locations and retinal structure

Purpose : Optical coherence tomography (OCT) two dimensional (2D) ganglion cell inner plexiform layer (GCIPL) thickness maps often reveal subtle abnormalities that might be washed out with summarized parameters (global or sectoral measurements). Also, the spatial pattern of GCIPL shows useful information to understand the extent and magnitude of localized damages. The purpose of this study was to predict next-visit 2D GCIPL thickness map based on the current and past GCIPL thickness maps. Methods : 346 glaucomatous eyes (191 subjects) with at least 5 visits with OCT tests were included in the study. GCIPL thickness maps were obtained using a clinical OCT (Cirrus HD-OCT, Zeiss, Dublin, CA; software version 9.5.1.13585; 200x200 macular cube scan). Since 83.2% of subjects were stable (average GCIPL change < 2um per year), we simulated progressing cases for diffuse damage pattern and hemifield damage pattern (superior vs. inferior hemifield damage was 50:50) (Figure 1 (c) and (d)). A deep learning based method, time-aware convolutional long short-term memory (TC-LSTM), was developed to handle irregular time intervals of longitudinal GCIPL thickness maps and predict the 5th GCIPL thickness map from the past 4 tests. The TC-LSTM model was compared with a conventional linear regression (LR) analysis. Mean square error (MSE, normalized to pixel intensity) and peak signal to noise ratio (PSNR) between predicted maps and ground truth maps were used to quantify the prediction quality (lower MSE and higher PSNR indicate better results). The Wilcoxon signed-rank test was used to compare TC-LSTM results and LR results. Results : TC-LSTM achieved lower MSE and higher PSNR compared to the LR model (MSE 0.00049 vs. 0.00061, p<0.001, and PSNR 34.45 vs. 32.52 dB, p=0.035). Subjective evaluation by 3 expert ophthalmologists showed that TC-LSTM model had closer representations of the ground truth maps than the LR model (Table 1, Figure 1). Conclusions : The next visit GCIPL thickness maps were successfully generated using TC-LSTM with higher accuracy compared to LR model both quantitatively and subjectively

Purpose : Gradient class activation maps (grad-CAM) generated by convolutional neural networks (CNN) have qualitatively indicated that these networks are able to identify important regions in OCT scans. Here, we quantitatively analyse these regions to improve our understanding of the CNN decision making process when detecting glaucoma in OCT volumes. Methods : A total of 1110 OCT (Cirrus HD-OCT, Zeiss, Dublin, Ca) scans from both eyes of 624 subjects (139 healthy and 485 glaucomatous patients (POAG)). An end-to-end 3D-CNN network was trained directly on 3D-volumes for glaucoma detection. Grad-CAM was implemented to highlight structures in the volumes that the network relied on. Grad-CAM heatmaps were generated for 3 different convolutional layers and quantitatively validated by occluding the regions with the highest grad-CAM weights (12.5% of original input volumes) and then evaluating the performance drop. Further, 8-retinal layers segmentation method was used to compute the average heatmap weights for each segmented layer separately, and used to identify the layers that were deemed as important for the task. Results : The model achieved an AUC of 0.97 for the test set (110 scans). Occlusion resulted in a 40% drop in performance (Fig.1). The RNFL and photoreceptors showed the highest median weights for grad-CAM heatmaps (0.1 and 0.2, respectively). The retinal pigment epithelium (RPE) and photoreceptors showed higher weights in the glaucomatous scans (Fig.2-a). RNFL had wider range of weights in healthy cases versus POAG ones. Analysis of the B-scans showed that central part around the optic disc (# 85-135) had the highest contribution to the network decision and the heatmap weights were much higher in glaucoma cases than healthy ones across all B-scans (Fig.2-b). Conclusions : The occlusion experiment indicates that the regions identified by the grad-CAMs are in fact pertinent to the glaucoma detection task. The increased emphasis on the photoreceptors in the glaucoma cases may be attributed to the atrophy in the superficial layers which in turn increased the brightness of this structure. This technique can be used to identify new biomarkers learned for other ocular diseases

Purpose : Open-angle glaucoma (OAG) is a leading cause of blindness globally, yet the relative contributions of genetic background and the environment to the development of this disease are unclear. As a first step in determining these contributions, we document posterior pole pathology and investigate the association between optic nerve head glaucomatous features, intraocular pressure (IOP), and demographic information, in an exceptionally large cohort of free-ranging rhesus macaques. Methods : We administered ophthalmologicl exams under sedation to 55 female and 54 male animals aged 0-21 years (mean age = 6.08; SD = 4.27). IOP was measured using TonoPen and TonoVet Plus tonometers and measurements adjusted using published calibration equations (McAllister et al Optom Vis Sci. 2018). Cup/Disk ratio (CDR) was calculated from clinical examination aided by optical coherence tomography (OCT; Bioptigen Envisu). Posterior pole pathology was documented using fundus imaging. Association between CDR and age, sex, and IOP was assessed using generalized linear mixed models (GLMM) and likelihood ratio tests (LRTs). Results : The mean (+/-standard deviation) IOP in the 218 eyes measured was 19.32 (+/-6.24) mmHg. Mean CDR was 0.37 (+/-0.15). We detected elevated IOP (> 22mmHg) in 78 eyes (36.62%) and CDR > 0.7 was detected in 13 eyes (5.96 %). CDR values were highly concordant in eyes of the same animal (CDR of left and right eyes within 0.2 for all animals. IOP was a significant predictor of CDR (p<0.001) in models that either included or excluded age, and animal sex. The best fitting model included only IOP as a predictor variable (AIC =-196.3). This model was significantly better than models containing age (AIC =-185.8, p = 0.004) or both age and sex (AIC =-183.4). Additional posterior pole pathology included pigmentary macular changes (8 eyes), macular scars (2 eyes), vessel tortuosity (19 eyes), and retinal hemorrhages (5 eyes). Conclusions : IOP is a significant predictor of CDR in this cohort. Age did not appear to correlate with CDR, but controlling for relatedness may further elucidate impacts of individual biology on OAG. Similarities between rhesus and human glaucomatous phenotypes and the presence of additional retinal pathology in our population may make these animals valuable in the study of other complex human disease, such as age-related macular degeneration

Purpose : In mild glaucoma, RNFL thinning and visual field (VF) loss are often localized, but structure-function modeling is impeded by variability due to individual eye anatomy. We perform high-resolution spatial correlations of RNFLT maps for each VF location to identify relevant areas and study further improvements by geometrically aligning RNFLT maps based on artery trajectories. Methods : In 419 SITA Standard 24-2 Humphrey VFs with at most mild glaucoma (mean deviation >=-3dB) with accompanying circumpapillary Cirrus HD-OCT RNFLT maps, we computed pixel-wise correlations (52 VF locations x 40401 pixels). We then performed an alignment operation, ensuring that the two major retinal arteries follow the same lines in all scans. We piecewise linearly approximated the trajectories of the arteries on 4 concentric circles around ONH (Fig. 1a), determined the necessary rotation for each pixel, and morphed the images accordingly (Fig. 1b). Results : For the pre-alignment RNFLT (correlation maps Fig. 2 top) we observed: (1) relatively high correlations (max 0.29); (2) most of the high-correlation regions are highly localized around the median trajectories of the major arteries at most VF locations, possibly due to the stacked character of the fiber bundles close to ONH, which impedes precise spatial mapping to the VF. This observation suggests general retinal vulnerability zones rather than highly VF location-specific areas as assumed by many previous structure-function models. Accordingly, morphing the RNFLT maps by aligning the eyespecific artery locations increased the maximal correlations on 25 of the 52 VF locations (Fig. 2 bottom, marked in green), particularly in nasal and inferior VF, with improvements of up to 0.1 (inferior arcuate region of VF). At many locations, aligned vulnerability areas become substantially more conspicuous (e.g. the location enlarged on the top left) and might have been missed without aligning. Conclusions : High-resolution structure-function correlations reveal retinal vulnerability zones in mild glaucoma. At many VF locations, these zones become better correlated with VF regions when RNFLT maps are aligned along the arteries. Specific attention to RNFL thinning in these zones in glaucoma suspects may improve the detection of initial VF loss glaucoma

Purpose : Oxygen concentration in retinal blood vessels (sO ) can be critical biomarkers for diabetic retinopathy and glaucoma, leading causes of blindness worldwide. We previously demonstrated sO2measurements in rodent and human retinas with spectroscopic visible-light optical coherence tomography (vis-OCT). However, reliable measurements of sO2in a clinical setting remains an open challenge due to constraints on light exposure, imaging time, patient motion, and variation in eye geometry. Spectral calibration to optimize sO2measurements under these non-ideal imaging conditions is needed. Here, we investigate, develop, and implement such calibration. Methods : We developed vis-OCT processing software to optimize sO2measurements in humans. First, we identified an optimal spectral range for spectral measurement in which sO2was most stable. Next, we developed methods to account for alterations induced by the imaging system and eye optics. Specifically, we accounted for depth-dependent variations in the measured spectrum, such as absorption contrast, spectrally-dependent roll-off, chromatic aberrations, and eye morphology. We then imaged the retinas of 12 healthy subjects aged 22 to 60 at Northwestern Medical Hospital in Chicago, IL, and Langone Medical Center in New York, NY. All imaging was approved by the respective IRBs and strictly adhered to the Declaration of Helsinki. Light exposure in the eye was no higher than 250 muW and imaging time was no longer than 5 s. We extracted sO2from vessels larger than 50 mum in diameter using an automated version of our vis-OCT processing software. Results : We measured the sO2in 89 vessels (53 arteries and 36 veins). We found the mean sO2in arteries was 97.70 +/-4.75 % in arteries and mean sO2in veins was 53.11 +/-6.85 %. Conclusions : We developed analytical methods for depth-dependent alterations to the measured spectrum in vis-OCT retinal oximetry. Our measurements yielded spectra that are highly consistent with those reported in literature, despite variations in imaging conditions. Our results indicate a clear path forward for clinical adoption of vis-OCT

Purpose : The presence of s zone peripapillary atrophy (PPA) has been associated with glaucoma. We performed a retrospective longitudinal study to evaluate s zone PPA area as a predictor for glaucomatous structural and functional progression. Methods : Subjects with glaucoma and >4 visits were included. Subjects had Humphrey visual field (Zeiss, Dublin, CA) testing, spectral-domain OCT (Cirrus HD-OCT; Zeiss) optic nerve head (ONH) and macula scans. s zone PPA was manually delineated on the baseline en face ONH scan as the area contiguous with the optic disc with the presence of hyper-and hyporeflectivity. Mixed effects linear models accounting for intra-subject correlation, follow-up time, scan's signal strength and ethnicity, were performed to determine if baseline PPA area was associated with glaucoma severity. Subsequent models incorporating the interaction term between time and baseline PPA area were performed to determine if baseline PPA area affected the rate of change in parameters of glaucoma over time. Results : 81 eyes (56 subjects) aged 62.8+/-14.1 years with an average follow-up time 3.9+/-1.3 years were analyzed. PPA was significantly associated with mean deviation (MD), visual field index (VFI), and inferior retinal nerve fiber layer (RNFL), (p=0.033, 0.038, and 0.034, respectively), but not with average RNFL, or macular ganglion cell inner plexiform layer (GCIPL) global and sectoral measurements and ONH parameters. No significant association was detected between s zone PPA area and the rate of progression for any parameter except for VFI (p =0.035). Conclusions : Although baseline s zone PPA area is associated with some indicators of glaucoma severity, it is not a significant predictor of the rate of glaucomatous progression (except for VFI)

Purpose : To develop a method for monitoring the functional deterioration of glaucoma patents using structural surrogates, we used machine learning algorithms to estimate visual field index (VFI) from OCT scans, and evaluated the accuracy of the progression rates calculated from the estimated VFI. Methods : Macular and ONH SDOCT scans (Cirrus HD-OCT, Zeiss, Dublin, CA; 200x200x1024 samplings over 6x6x2mm, downsampled to 64x64x128 voxels) were acquired from both eyes of 1,678 healthy participants, glaucoma suspects, and glaucoma patients over multiple visits (range: 1-14, median=3), forming a dataset of 10,172 pairs of macular+ONH scans. Automated perimetry (Humphrey visual field, SITA 24-2) tests were administered at each visit. Two models were trained to estimate the measured VFI from a pair of macular and ONH scans: the first ("classic model") was a non-linear regression model (multi-layer perceptron) based on 47 thickness measures of retinal layers, while the other ("CNN") was a 5-layer convolutional neural network, trained to learn 3D features in the OCT scans. For both models, MSE was minimized in 5-fold cross-validation, using 80%:10%:10% of the dataset as training, validation and test sets. Data from the same participant were not split across the three sets. For data in the test sets, VFI's for eyes with more than N=3,4,5 visits were estimated for individual visits, and the slopes were calculated using linear regression across N consecutive visits. Median absolute error (MAE) was used to quantify estimation accuracy. Results : For estimating VFI at single visits, the CNN achieved significant lower MAE (2.6+/-0.28; mean and s.d.) than the classic model (2.9+/-0.45). For estimating slopes across 5 visits, the MAE of the CNN (0.73+/-0.12/year) was also lower than the classic model (0.82+/-0.23/year). The errors depended on the measured VFI of the first visit, and on the true slope (Fig. 1). Increasing the number of visits decreased the errors (N=3.6, MAE=1.38/yr, 0.99/yr, 0.73/yr, and 0.63yr) Conclusions : The feature-agnostic CNN was better at estimating VFI and visual field progression rates than the regression method based on thickness measures. Structure-tofunction estimation using neural networks is a promising method for monitoring the visual functions of glaucoma patients

Observational studies in glaucoma patients can provide important evidence on treatment effects, especially for combination therapies which are often used in reality. But the success relies on the reduction of selection bias through methods such as propensity score (PS) weighting. The objective of this study was to assess the effects of five glaucoma treatments (medication, laser, non-laser surgery (NLS), laser + medication, and NLS + medication) on 1-year intraocular pressure (IOP) change. Data were collected from 90 glaucoma subjects who underwent a single laser, or NLS intervention, and/or took the same medication for at least 6 months, and had IOP measures before the treatment and 12-months after. Baseline IOP was significantly different across groups (p = 0.007) and this unbalance was successfully corrected by the PS weighting (p = 0.81). All groups showed statistically significant PS-weighted IOP reductions, with the largest reduction in NLS group (-6.78 mmHg). Baseline IOP significantly interacted with treatments (p = 0.03), and at high baseline IOP medication was less effective than other treatments. Our findings showed that the 1-year IOP reduction differed across treatment groups and was dependent on baseline IOP. The use of PS-weighted methods reduced treatment selection bias at baseline and allowed valid assessment of the treatment effect in an observational study.

Glaucoma is the world's leading cause of irreversible blindness, and falls are a major public health concern in glaucoma patients. Although recent evidence suggests the involvements of the brain toward advanced glaucoma stages, the early brain changes and their clinical and behavioral consequences remain poorly described. This study aims to determine how glaucoma may impair the brain structurally and functionally within and beyond the visual pathway in the early stages, and whether these changes can explain visuomotor impairments in glaucoma. Using multi-parametric magnetic resonance imaging, glaucoma patients presented compromised white matter integrity along the central visual pathway and around the supramarginal gyrus, as well as reduced functional connectivity between the supramarginal gyrus and the visual occipital and superior sensorimotor areas when compared to healthy controls. Furthermore, decreased functional connectivity between the supramarginal gyrus and the visual brain network may negatively impact postural control measured with dynamic posturography in glaucoma patients. Taken together, this study demonstrates that widespread structural and functional brain reorganization is taking place in areas associated with visuomotor coordination in early glaucoma. These results implicate an important central mechanism by which glaucoma patients may be susceptible to visual impairments and increased risk of falls.