An interpretable and interactive deep learning algorithm for a clinically applicable retinal fundus diagnosis system by modelling finding-disease relationship.

The identification of abnormal findings manifested in retinal fundus images and diagnosis of ophthalmic diseases are essential to the management of potentially vision-threatening eye conditions. Recently, deep learning-based computer-aided diagnosis systems (CADs) have demonstrated their potential to reduce reading time and discrepancy amongst readers. However, the obscure reasoning of deep neural networks (DNNs) has been the leading cause to reluctance in its clinical use as CAD systems. Here, we present a novel architectural and algorithmic design of DNNs to comprehensively identify 15 abnormal retinal findings and diagnose 8 major ophthalmic diseases from macula-centered fundus images with the accuracy comparable to experts. We then define a notion of counterfactual attribution ratio (CAR) which luminates the system's diagnostic reasoning, representing how each abnormal finding contributed to its diagnostic prediction. By using CAR, we show that both quantitative and qualitative interpretation and interactive adjustment of the CAD result can be achieved. A comparison of the model's CAR with experts' finding-disease diagnosis correlation confirms that the proposed model identifies the relationship between findings and diseases similarly as ophthalmologists do.

Sequential structural and functional change in geographic atrophy on multimodal imaging in non-exudative age-related macular degeneration.

PURPOSE: To investigate the temporal order of photoreceptor atrophy, retinal pigment epithelium (RPE) atrophy and visual acuity loss in patients with center-involving geographic atrophy (GA) in non-exudative age-related macular degeneration (neAMD). METHODS: Forty eyes of 25 consecutive patients who eventually developed center-involving GA were investigated. Fundus autofluorescence (FAF) and infrared image coupled optical coherence tomography (OCT) were acquired at each visit. Development of RPE atrophy and photoreceptor atrophy was defined as abnormal hyper/hypo-fluorescence on FAF and photoreceptor loss on OCT over 50% of the vertical or horizontal diameters of the center 1 mm circle, respectively. Visual acuity loss was defined as worsening of more than 0.2 logMAR compared to baseline. Kaplan-Meier analyses was performed to compare the sequential order of these three events. RESULTS: Mean age was 72.72 ± 8.63 years, and follow-up duration was 27.36 ± 17.22 months, with an average number of visits of 3.04 ± 1.54 during follow-up. GA progressed from photoreceptor atrophy on OCT, RPE atrophy on FAF, and then to vision loss (p < 0.001). The median survival time of photoreceptors preceded that of visual acuity by 16.3 months, and the median survival time of RPE preceded that of visual acuity by 7.0 months. At baseline, majority of eyes showed drusen only (57.5%), while the most common feature was incomplete RPE and outer retinal atrophy at 3-year follow-up (40.4%). CONCLUSION: In the progression of center-involving GA, photoreceptor atrophy on OCT and RPE atrophy on FAF precedes visual decline, and can act as biomarkers predicting future visual decline within the following years.

Comparison of Optical Coherence Tomography Biomarkers between Bevacizumab Good Responders and Nonresponders Who were Switched to Dexamethasone Implant in Diabetic Macular Edema.

PURPOSE: To compare volumetric optical coherence tomography (OCT) biomarkers in bevacizumab responsive and bevacizumab refractory diabetic macular edema (DME) patients switched to the dexamethasone implant to ultimately identify possible prognostic indicators. METHODS: Retrospective analysis of DME patients treated with bevacizumab were done. Patients were divided into those who showed response to bevacizumab (bevacizumab only group) and others who were switched to the dexamethasone implant due to lack of response to bevacizumab (switching group). Volumetric OCT biomarkers such as central macular thickness (CMT), inner and outer cystoid macular edema (CME) volume, serous retinal detachment (SRD) volume, retinal volume (CME + SRD volume) within the 6-mm Early Treatment of Diabetic Retinopathy Study circle were calculated. OCT biomarkers were followed up throughout treatment. RESULTS: Among total of 144 eyes, 113 patients were included in the bevacizumab only group and 31 patients were included in the switching group. Compared to the bevacizumab only group, the switching group showed higher baseline CMT (558.00 ± 209.60 µm vs. 454.96 ± 125.88 µm, p = 0.003), larger inner CME (6.02 ± 1.43 mm3 vs. 5.12 ± 0.87 mm3, p = 0.004) and SRD volume (0.32 ± 0.40 mm3 vs. 0.11 ± 0.09 mm3, p = 0.015) and higher proportion of patients with SRD (58.06% vs. 31.86%, p = 0.008). In the switching group, CMT, inner CME and SRD volume all showed significant reduction after switching to the dexamethasone implant. CONCLUSIONS: DME with large SRD and inner nuclear layer edema volume may be more effectively treated with the dexamethasone implant than bevacizumab.

Assessment of Risk Factors Affecting Refractive Outcomes after Phacovitrectomy for Epiretinal Membrane.

PURPOSE: To investigate factors associated with refractive outcomes after phacovitrectomy for epiretinal membrane (ERM). METHODS: Retrospective review of patients undergoing phacovitrectomy for ERM was done. The main outcome measure was predictive refraction error (PE), defined as observed refraction error - target refraction error, calculated by the SRK/T, Haigis, and SRK II formulae. PE was measured at postoperative 1, 3, and 6 months. Simple and multiple linear regression analysis were used to evaluate factors associated with PE. RESULTS: A total of 53 eyes of 53 patients were included. The mean PEs at postoperative 1, 3, and 6 months were all negative, implying myopic shift in all patients regardless of the intraocular lens formula used. Haigis formula showed the least myopic shift among the three formulae (p = 0.001, Friedman test). There was no significant difference in PE depending on preoperative central macular thickness (CMT) in subgroup analysis. On stepwise multiple linear regression analysis, ERM etiology (ß = 0.759, p = 0.004, SRK/T formula; ß = 0.733, p = 0.008, Haigis formula; ß = 0.933, p < 0.001, SRK II formula), preoperative anterior chamber depth (ß = -0.662, p = 0.013, Haigis formula; ß = -0.747, p = 0.003, SRK II formula), and decrease of CMT (ß = -0.003, p = 0.025, SRK/T formula) were significantly associated with PE at postoperative 6 months. CONCLUSIONS: Myopic shift in PE was observed after combined phacovitrectomy for epiretinal membrane. ERM etiology, preoperative anterior chamber depth, and decrease of CMT were significantly associated with PE at postoperative 6 months. There was no difference in PE after surgery between the two groups defined by CMT (≥500 and <500 µm).

Comparative risk of blindness and vision-threatening ocular comorbidities in patients with Behçet's disease versus the general population.

OBJECTIVE: To compare the risk of blindness and vision-threatening ocular comorbidities in patients with Behçet's disease (BD) vs the general population. METHODS: Using 2002-2017 Korea National Health Insurance Service database, we did a population-based cohort study comparing newly diagnosed BD patients and age- and sex-matched non-BD controls at a 1:5 ratio. The primary outcome was blindness, defined as a best-corrected visual acuity of ≤20/500 in the better-seeing eye. Secondary outcomes were vision-threatening ocular comorbidities (cataract, glaucoma and retinal disorders) that require surgical interventions and incident uveitis. Cox proportional hazards models were used to estimate hazard ratios (HRs) and 95% CIs. We performed subgroup analyses by sex and BD diagnosis age. RESULTS: We included 31 228 BD patients and 156 140 controls. During a follow-up of 9.39 years, the incidence rate of blindness per 1000 person-years was 0.24 in BD and 0.02 in controls with an HR of 10.73 (95% CI 7.10, 16.22). The HR for secondary outcomes was 2.06 (95% CI 1.98, 2.15) for cataract surgery, 5.43 (4.57, 6.45) for glaucoma surgery and 2.71 (2.39, 3.07) for retinal surgery. The HR of incident uveitis was 6.19 (95% CI 5.83, 6.58). Males suffered a disproportionately higher risk of blindness than females due to greater severity rather than a lower incidence of uveitis. The risk of uveitis and blindness decreased as BD diagnosis age increased. CONCLUSIONS: In this large population-based cohort study, BD patients compared with the general population have a 10.73-fold risk of blindness in 10 years and also a substantially higher risk of diverse ocular comorbidities that pose potential threats to vision.

Evaluation of changes for sebum, skin pore, texture, and redness before and after sleep in oily and nonoily skin.

BACKGROUND: People whose skin type is oily have experienced an esthetic and hygienic discomfort due to the excessive secretion of the sebum during the day and night time, and therefore sebum control is required. In this study, we aimed to find out whether the skin status between the oily and nonoily skin indicates a significant difference before and after sleep. MATERIALS AND METHODS: Forty Korean males and females whose skin type was oily or nonoily participated in this study. To investigate the difference of the skin between oily and nonoily skin before and after sleep, we measured the sebum, skin pore, texture, and redness on their cheek at baseline and after 4-h sleep. Moreover, the significant level was determined at p < 0.05. RESULTS: Parameters of sebum and skin pore significantly increased after 4-h sleep compared with baseline in the oily and nonoily skin (p < 0.05). Moreover, the increment of sebum and pore parameters in the oily skin was significantly higher than those in the nonoily skin (p < 0.05). In the case of skin texture and redness, parameters of them were significantly changed after 4-h sleep compared with baseline only in the oily skin (p < 0.05), and there was no significant difference among groups. CONCLUSIONS: We found that the change rates of the sebum secretion and skin pore in oily skin were significantly higher than those in nonoily skin after 4-h sleep. These results suggest the necessity of the skin care depending on the skin type before sleeping.

Peripapillary retinal nerve fiber layer thinning in patients with progressive supranuclear palsy.

OBJECTIVES: To investigate peripapillary retinal nerve fiber layer (pRNFL) changes in patients with progressive supranuclear palsy (PSP). METHODS: We included 21 PSP patients (36 eyes) who underwent peripapillary optical coherence tomography (OCT) scans at 2.5 ± 1.3 years of disease, without ophthalmologic co-morbidities. We compared pRNFL thicknesses in PSP eyes with age-matched 22 controls (22 eyes) using generalized estimating equation model adjusting for intra-subject inter-eye correlations, age and sex. We also analyzed the correlation between the pRNFL thickness and clinical severity using Spearman's correlation. In twelve PSP patients with 3 T brain MRI volumetric scan within 1 year of OCT exam, we investigated the correlation between the pRNFL thickness and brain atrophy using Pearson's correlation. RESULTS: PSP patients had global pRNFL thinning compared to controls (beta = - 6.436, p = 0.025). Global pRNFL thickness correlated with Hoehn & Yahr stages (r = - 0.487, p = 0.025), and nasal pRNFL thinning showed a trend of correlation (uncorrected p < 0.05). Exploratory correlation analysis between global pRNFL thickness and nonmotor items in the PSP rating scale showed a trend toward association with sleep disturbances (uncorrected p = 0.008) and urinary incontinence (uncorrected p = 0.031), although not significant after Bonferroni correction (all 28 items). The patients had significant atrophy in the posterior cingulate cortex, third ventricle, pallidum, and midbrain with reduced midbrain-to-pons ratio, but no correlation was found between pRNFL thickness and brain volumes. CONCLUSION: The pRNFL seems to be affected in PSP, which is more severe with advanced disease stages. Retinal investigation in a larger longitudinal cohort would help elucidate the pathophysiological role of retinal thinning in PSP.

Embryonic stem cell-derived photoreceptor precursor cells differentiated by coculture with RPE cells.

Purpose: To describe the derivation of photoreceptor precursor cells from human embryonic stem cells by coculture with RPE cells. Methods: Human embryonic stem cells were induced to differentiate into neural precursor cells and then cocultured with RPE cells to obtain cells showing retinal photoreceptor features. Immunofluorescent staining, reverse transcription-PCR (RT-PCR), and microarray analysis were performed to identify photoreceptor markers, and a cGMP assay was used for in vitro functional analysis. After subretinal injection in rat animal models, retinal function was determined with electroretinography and optokinetic response detection, and immunofluorescent staining was performed to assess the survival of the injected cells. Results: Cocultured cells were positive for rhodopsin, red and blue opsin, recoverin, and phosphodiesterase 6 beta on immunofluorescent staining and RT-PCR. Serial detection of stem cell-, neural precursor-, and photoreceptor-specific markers was noted in each stage of differentiation with microarray analysis. Increased cGMP hydrolysis in light-exposed conditions compared to that in dark conditions was observed. After the subretinal injection in the rats, preservation of optokinetic responses was noted up to 20 weeks, while electroretinographic response decreased. Survival of the injected cells was confirmed with positive immunofluorescence staining of human markers at 8 weeks. Conclusions: Cells showed photoreceptor-specific features when stem cell-derived neurogenic precursors were cocultured with RPE cells.

Microbial diagnosis of endophthalmitis using nanopore amplicon sequencing.

OBJECTIVES: We investigated whether nanopore amplicon sequencing of aqueous humor was capable of rapid pathogen identification in infectious endophthalmitis. METHODS: 5 cases of culture-positive bacterial endophthalmitis and 3 cases of fungal endophthalmitis (1 culture-positive and 2 presumed) were included. DNA was extracted from the aqueous humor and vitreous specimen, and PCR of bacterial rDNA (16S) and fungal rDNA (ITS1 and D1/2/3) was performed. Then, nanopore amplicon sequencing was performed for 2 h. The results of amplicon sequencing were compared to those of conventional culture studies. RESULTS: In all cases, pathogens were identified by amplicon sequencing of aqueous humor specimens. In 3 cases of bacterial endophthalmitis, the identified microbes were confirmed by culture studies of both aqueous humor and vitreous specimens. In 2 cases of bacterial and 1 case of fungal endophthalmitis, the identified pathogens were confirmed only by culture studies of vitreous specimens. In all cases, amplicon sequencing identified pathogen in a shorter turnaround time than culture studies. In 2 cases with negative culture results, amplicon sequencing of aqueous humor identified fungal pathogens. CONCLUSIONS: Our data demonstrates the potential of amplicon nanopore sequencing using aqueous humor to enable rapid, sensitive and less invasive microbial diagnosis of endophthalmitis.

Efficacy of bevacizumab for vitreous haemorrhage in proliferative diabetic retinopathy with prior complete panretinal photocoagulation.

PURPOSE: To investigate the efficacy of intravitreal bevacizumab injections (IVBs) for vitreous haemorrhage (VH) in proliferative diabetic retinopathy (PDR) with prior complete panretinal photocoagulation (PRP). METHODS: A multicentre cohort study of eyes with new VH in PDR after documented previous complete PRP was performed. Eyes were grouped according to IVB treatment at baseline, and cumulative rate of vitrectomy and spontaneous clear-up rate were compared as the main outcome. Eyes requiring vitrectomy within 1 month, or with tractional retinal detachment (TRD), or with spontaneous clear-up within 1 month, were excluded. RESULTS: In total, 44 eyes with IVB and 92 control eyes without IVB were followed up to 20.1 months. Cumulative probability of vitrectomy was lower in the IVB group at 12 months (0.16 vs 0.42, IVB vs controls), and throughout the follow-up period (p = 0.005). Cumulative probability of spontaneous clear-up was higher in the IVB group at 12 months (0.81 vs 0.68, IVB vs controls), and throughout the follow-up period (p = 0.013). Best-corrected visual acuity (BCVA) at 1 month after onset of VH was significantly better in the IVB group (0.513 vs 0.942 logarithm of the minimal angle of resolution, p = 0.002); however, the difference of BCVA lost significance with further follow-up. IVB treatment was the only factor significantly associated with vitrectomy risk on multivariate analysis (p = 0.047, hazard ratio 0.506). CONCLUSION: In VH after prior complete PRP, IVB was effective in decreasing vitrectomy requirement, although overall visual benefit was short-term. IVB can be considered to defer vitrectomy in PDR VH eyes with prior complete PRP and no TRD.

Explaining the Rationale of Deep Learning Glaucoma Decisions with Adversarial Examples.

PURPOSE: To illustrate what is inside the so-called black box of deep learning models (DLMs) so that clinicians can have greater confidence in the conclusions of artificial intelligence by evaluating adversarial explanation on its ability to explain the rationale of DLM decisions for glaucoma and glaucoma-related findings. Adversarial explanation generates adversarial examples (AEs), or images that have been changed to gain or lose pathologic characteristic-specific traits, to explain the DLM's rationale. DESIGN: Evaluation of explanation methods for DLMs. PARTICIPANTS: Health screening participants (n = 1653) at the Seoul National University Hospital Health Promotion Center, Seoul, Republic of Korea. METHODS: We trained DLMs for referable glaucoma (RG), increased cup-to-disc ratio (ICDR), disc rim narrowing (DRN), and retinal nerve fiber layer defect (RNFLD) using 6430 retinal fundus images. Surveys consisting of explanations using AE and gradient-weighted class activation mapping (GradCAM), a conventional heatmap-based explanation method, were generated for 400 pathologic and healthy patient eyes. For each method, board-trained glaucoma specialists rated location explainability, the ability to pinpoint decision-relevant areas in the image, and rationale explainability, the ability to inform the user on the model's reasoning for the decision based on pathologic features. Scores were compared by paired Wilcoxon signed-rank test. MAIN OUTCOME MEASURES: Area under the receiver operating characteristic curve (AUC), sensitivities, and specificities of DLMs; visualization of clinical pathologic changes of AEs; and survey scores for locational and rationale explainability. RESULTS: The AUCs were 0.90, 0.99, 0.95, and 0.79 and sensitivities were 0.79, 1.00, 0.82, and 0.55 at 0.90 specificity for RG, ICDR, DRN, and RNFLD DLMs, respectively. Generated AEs showed valid clinical feature changes, and survey results for location explainability were 3.94 ± 1.33 and 2.55 ± 1.24 using AEs and GradCAMs, respectively, of a possible maximum score of 5 points. The scores for rationale explainability were 3.97 ± 1.31 and 2.10 ± 1.25 for AEs and GradCAM, respectively. Adversarial example provided significantly better explainability than GradCAM. CONCLUSIONS: Adversarial explanation increased the explainability over GradCAM, a conventional heatmap-based explanation method. Adversarial explanation may help medical professionals understand more clearly the rationale of DLMs when using them for clinical decisions.

Predicting High Coronary Artery Calcium Score From Retinal Fundus Images With Deep Learning Algorithms.

Purpose: To evaluate high accumulation of coronary artery calcium (CAC) from retinal fundus images with deep learning technologies as an inexpensive and radiation-free screening method. Methods: Individuals who underwent bilateral retinal fundus imaging and CAC score (CACS) evaluation from coronary computed tomography scans on the same day were identified. With this database, performances of deep learning algorithms (inception-v3) to distinguish high CACS from CACS of 0 were evaluated at various thresholds for high CACS. Vessel-inpainted and fovea-inpainted images were also used as input to investigate areas of interest in determining CACS. Results: A total of 44,184 images from 20,130 individuals were included. A deep learning algorithm for discrimination of no CAC from CACS >100 achieved area under receiver operating curve (AUROC) of 82.3% (79.5%-85.0%) and 83.2% (80.2%-86.3%) using unilateral and bilateral fundus images, respectively, under a 5-fold cross validation setting. AUROC increased as the criterion for high CACS was increased, showing a plateau at 100 and losing significant improvement thereafter. AUROC decreased when fovea was inpainted and decreased further when vessels were inpainted, whereas AUROC increased when bilateral images were used as input. Conclusions: Visual patterns of retinal fundus images in subjects with CACS > 100 could be recognized by deep learning algorithms compared with those with no CAC. Exploiting bilateral images improves discrimination performance, and ablation studies removing retinal vasculature or fovea suggest that recognizable patterns reside mainly in these areas. Translational Relevance: Retinal fundus images can be used by deep learning algorithms for prediction of high CACS.

Association of Cardiovascular Mortality and Deep Learning-Funduscopic Atherosclerosis Score derived from Retinal Fundus Images.

PURPOSE: The prediction of atherosclerosis using retinal fundus images and deep learning has not been shown possible. The purpose of this study was to develop a deep learning model which predicted atherosclerosis by using retinal fundus images and to verify its clinical implications by conducting a retrospective cohort analysis. DESIGN: Retrospective cohort study. METHODS: The database at the Health Promotion Center of Seoul National University Hospital (HPC-SNUH) was used. The deep learning model was trained using 15,408 images to predict carotid artery atherosclerosis, which was named the deep-learning funduscopic atherosclerosis score (DL-FAS). A retrospective cohort was constructed of participants 30-80 years old who had completed elective health examinations at HPC-SNUH. Using DL-FAS as the main exposure, participants were followed for the primary outcome of death due to CVD until Dec. 31, 2017. RESULTS: For predicting carotid artery atherosclerosis among subjects, the model achieved an area under receiver operating curve (AUROC) and area under the precision-recall curve (AUPRC), accuracy, sensitivity, specificity, positive and negative predictive values of 0.713, 0.569, 0.583, 0.891, 0.404, 0.465, and 0.865 respectively. The cohort consisted of 32,227 participants, 78 cardiovascular disease (CVD) deaths, and 7.6-year median follow-up visits. Those with DL-FAS greater than 0.66 had an increased risk of CVD deaths compared to those with DL-FAS <0.33 (hazard ratio: 8.33; 95% confidence interval [CI], 3.16-24.7). Risk association was significant among intermediate and high Framingham risk score (FRS) subgroups. The DL-FAS improved the concordance by 0.0266 (95% CI, 0.0043-0.0489) over the FRS-only model. The relative integrated discrimination index was 20.45% and net reclassification index was 29.5%. CONCLUSIONS: A deep learning model was developed which could predict atherosclerosis from retinal fundus images. The resulting DL-FAS was an independent predictor of CVD deaths when adjusted for FRS and added predictive value over FRS.

Contrast sensitivity impairment in drug-naïve Parkinson's disease patients associates with early cognitive decline.

OBJECTIVES: To investigate the contrast sensitivity function in drug-naïve Parkinson's disease (PD) patients and its predictive value with longitudinal follow-up data. METHODS: We included newly diagnosed non-demented PD patients who performed contrast sensitivity test between 2013 and 2014. Contrast sensitivity function at drug-naïve state in PD patients was compared with age-matched normal control data of our center. Correlation between contrast sensitivity function and parkinsonian motor and non-motor features including the Mini-Mental State Exam (MMSE) score at the time of diagnosis were analyzed by linear regression. With longitudinal follow-up data after initiating anti-parkinsonian therapy, the risk conferred on subsequent visual hallucinations and cognitive impairment requiring anti-dementia drugs was analyzed by dichotomizing PD group based on the initial contrast sensitivity function. RESULTS: Forty-eight patients were finally included, and mean follow-up periods were 43 months. Contrast sensitivity function in drug-naïve PD patients was significantly worse than controls. Contrast sensitivity function correlated with sleep disturbance (p = 0.001) and global cognitive status reflected by the MMSE score (p = 0.020). It also associated with further decline in the MMSE during the follow-ups (p = 0.029). Patients with below average contrast sensitivity function at the time of diagnosis showed higher risk of cognitive decline requiring anti-dementia drugs (adjusted odds ratio = 4.68, p = 0.04) and of visual hallucinations (adjusted odds ratio = 12.54, p = 0.04) than those above average function during the follow-up. CONCLUSION: Contrast sensitivity impairment in drug-naïve PD patients associates with clinical demand for therapeutic intervention of cognitive decline as well as development of visual hallucinations in the early course of the disease.

Development and Validation of Deep Learning Models for Screening Multiple Abnormal Findings in Retinal Fundus Images.

PURPOSE: To develop and evaluate deep learning models that screen multiple abnormal findings in retinal fundus images. DESIGN: Cross-sectional study. PARTICIPANTS: For the development and testing of deep learning models, 309 786 readings from 103 262 images were used. Two additional external datasets (the Indian Diabetic Retinopathy Image Dataset and e-ophtha) were used for testing. A third external dataset (Messidor) was used for comparison of the models with human experts. METHODS: Macula-centered retinal fundus images from the Seoul National University Bundang Hospital Retina Image Archive, obtained at the health screening center and ophthalmology outpatient clinic at Seoul National University Bundang Hospital, were assessed for 12 major findings (hemorrhage, hard exudate, cotton-wool patch, drusen, membrane, macular hole, myelinated nerve fiber, chorioretinal atrophy or scar, any vascular abnormality, retinal nerve fiber layer defect, glaucomatous disc change, and nonglaucomatous disc change) with their regional information using deep learning algorithms. MAIN OUTCOME MEASURES: Area under the receiver operating characteristic curve and sensitivity and specificity of the deep learning algorithms at the highest harmonic mean were evaluated and compared with the performance of retina specialists, and visualization of the lesions was qualitatively analyzed. RESULTS: Areas under the receiver operating characteristic curves for all findings were high at 96.2% to 99.9% when tested in the in-house dataset. Lesion heatmaps highlight salient regions effectively in various findings. Areas under the receiver operating characteristic curves for diabetic retinopathy-related findings tested in the Indian Diabetic Retinopathy Image Dataset and e-ophtha dataset were 94.7% to 98.0%. The model demonstrated a performance that rivaled that of human experts, especially in the detection of hemorrhage, hard exudate, membrane, macular hole, myelinated nerve fiber, and glaucomatous disc change. CONCLUSIONS: Our deep learning algorithms with region guidance showed reliable performance for detection of multiple findings in macula-centered retinal fundus images. These interpretable, as well as reliable, classification outputs open the possibility for clinical use as an automated screening system for retinal fundus images.

Retina thickness as a marker of neurodegeneration in prodromal lewy body disease.

OBJECTIVES: We investigated retinal change and its relationship with neurodegeneration markers in a prodromal Parkinson cohort. METHODS: A total of 30 patients with idiopathic rapid eye movement sleep behavior disorder were recruited. Participants underwent olfactory testing, macular optical coherence tomography, microperimetry, contrast sensitivity test, and brain N-(3-[18 F]fluoropropyl)-2-carbomethoxy-3-(4-iodophenyl) nortropane positron emission tomography. We measured the ganglion cell complex thicknesses and investigated its correlation with olfactory function and striatal dopamine transporter availability. A linear mixed-effect model was applied with adjustment for multiple comparisons. RESULTS: The parafoveal ganglion-cell-complex thickness in this cohort lay between our healthy control and drug-naïve Parkinson's disease group data. Idiopathic rapid eye movement sleep behavior disorder patients also had contrast sensitivity impairment as in Parkinson's disease with a nonsignificant change in macular sensitivities. Macular ganglion cell complex thickness correlated with olfactory scores and with striatal dopamine transporter availabilities. CONCLUSIONS: Macular ganglion cell complex thinning may be a marker of neurodegeneration in prodromal Parkinson's disease. © 2019 International Parkinson and Movement Disorder Society.

Intraocular Pressure (IOP) Change and Frequency of IOP Spike After Cataract Surgery in Normal-tension Glaucoma: A Case-Control Study.

PURPOSE: To evaluate the early-postoperative intraocular pressure (IOP) changes and frequency of IOP spike after cataract surgery in normal-tension glaucoma (NTG) eyes and to compare them with those of nonglaucomatous eyes. MATERIALS AND METHODS: This was a case-control study. We reviewed the medical records of patients who had undergone cataract surgery. One-to-one (1:1) case matching was performed, each matched set consisting of 1 NTG and 1 nonglaucomatous eye as the control eye. Comparisons between those 2 groups were performed. Specifically, IOP was measured preoperatively and 1 day, 1, 2, and 4 weeks postoperatively. The incidence of IOP spike was evaluated according to the criteria of IOP higher than 21, 25, 30 mm Hg, higher than 5, 10 mm Hg over baseline IOP, and 50% above preoperative IOP. RESULTS: A total of 298 eyes (149 NTG eyes and 149 control eyes) were enrolled. Preoperative IOP did not show any significant difference between the 2 groups (P=0.687). IOP significantly decreased in both the NTG and control groups [P<0.001, repeated-measures analysis of variance (ANOVA)]. No significant differences in this regard were detected between the 2 groups (P=0.618, repeated-measures ANOVA). When IOP spike was evaluated according to the 6 criteria, the frequencies were <3% at any timepoint. There was no significant difference between the 2 groups. CONCLUSIONS: In both the NTG and control groups, IOP gradually decreased in the early-postoperative period after cataract surgery, and there was no significant difference between the 2 groups. The frequency of IOP spike was <5% in both the NTG group and the control group.

A Novel Fundus Image Reading Tool for Efficient Generation of a Multi-dimensional Categorical Image Database for Machine Learning Algorithm Training.

BACKGROUND: We described a novel multi-step retinal fundus image reading system for providing high-quality large data for machine learning algorithms, and assessed the grader variability in the large-scale dataset generated with this system. METHODS: A 5-step retinal fundus image reading tool was developed that rates image quality, presence of abnormality, findings with location information, diagnoses, and clinical significance. Each image was evaluated by 3 different graders. Agreements among graders for each decision were evaluated. RESULTS: The 234,242 readings of 79,458 images were collected from 55 licensed ophthalmologists during 6 months. The 34,364 images were graded as abnormal by at-least one rater. Of these, all three raters agreed in 46.6% in abnormality, while 69.9% of the images were rated as abnormal by two or more raters. Agreement rate of at-least two raters on a certain finding was 26.7%-65.2%, and complete agreement rate of all-three raters was 5.7%-43.3%. As for diagnoses, agreement of at-least two raters was 35.6%-65.6%, and complete agreement rate was 11.0%-40.0%. Agreement of findings and diagnoses were higher when restricted to images with prior complete agreement on abnormality. Retinal/glaucoma specialists showed higher agreements on findings and diagnoses of their corresponding subspecialties. CONCLUSION: This novel reading tool for retinal fundus images generated a large-scale dataset with high level of information, which can be utilized in future development of machine learning-based algorithms for automated identification of abnormal conditions and clinical decision supporting system. These results emphasize the importance of addressing grader variability in algorithm developments.