The BONSAI (Brain and Optic Nerve Study with Artificial Intelligence) deep learning system can accurately identify pediatric papilledema on standard ocular fundus photographs.

BACKGROUND: Pediatric papilledema often reflects an underlying severe neurologic disorder and may be difficult to appreciate, especially in young children. Ocular fundus photographs are easy to obtain even in young children and in nonophthalmology settings. The aim of our study was to ascertain whether an improved deep-learning system (DLS), previously validated in adults, can accurately identify papilledema and other optic disk abnormalities in children. METHODS: The DLS was tested on mydriatic fundus photographs obtained in a multiethnic pediatric population (<17 years) from three centers (Atlanta-USA; Bucharest-Romania; Singapore). The DLS's multiclass classification accuracy (ie, normal optic disk, papilledema, disks with other abnormality) was calculated, and the DLS's performance to specifically detect papilledema and normal disks was evaluated in a one-vs-rest strategy using the AUC, sensitivity and specificity, with reference to expert neuro-ophthalmologists. RESULTS: External testing was performed on 898 fundus photographs: 447 patients; mean age, 10.33 (231 patients ≤10 years of age; 216, 11-16 years); 558 normal disks, 254 papilledema, 86 other disk abnormalities. Overall multiclass accuracy of the DLS was 89.6% (range, 87.8%-91.6%). The DLS successfully distinguished "normal" from "abnormal" optic disks (AUC 0.99 [0.98-0.99]; sensitivity, 87.3% [84.9%-89.8%]; specificity, 98.5% [97.6%-99.6%]), and "papilledema" from "normal and other" (AUC 0.99 [0.98-1.0]; sensitivity, 98.0% [96.8%-99.4%]; specificity, 94.1% (92.4%-95.9%)]. CONCLUSIONS: Our DLS reliably distinguished papilledema from normal optic disks and other disk abnormalities in children, suggesting it could be utilized as a diagnostic aid for the assessment of optic nerve head appearance in the pediatric age group.

Application of a Deep Learning System to Detect Papilledema on Nonmydriatic Ocular Fundus Photographs in an Emergency Department.

PURPOSE: The Fundus photography vs Ophthalmoscopy Trial Outcomes in the Emergency Department (FOTO-ED) studies showed that ED providers poorly recognized funduscopic findings in patients in the ED. We tested a modified version of the Brain and Optic Nerve Study Artificial Intelligence (BONSAI) deep learning system on nonmydriatic fundus photographs from the FOTO-ED studies to determine if the deep learning system could have improved the detection of papilledema had it been available to ED providers as a real-time diagnostic aid. DESIGN: Retrospective secondary analysis of a cohort of patients included in the FOTO-ED studies. METHODS: The testing data set included 1608 photographs obtained from 828 patients in the FOTO-ED studies. Photographs were reclassified according to the optic disc classification system used by the deep learning system ("normal optic discs," "papilledema," and "other optic disc abnormalities"). The system's performance was evaluated by calculating the area under the receiver operating characteristic curve (AUC), sensitivity, and specificity using a 1-vs-rest strategy, with reference to expert neuro-ophthalmologists. RESULTS: The BONSAI deep learning system successfully distinguished normal from abnormal optic discs (AUC 0.92 [95% confidence interval {CI} 0.90-0.93]; sensitivity 75.6% [73.7%-77.5%] and specificity 89.6% [86.3%-92.8%]), and papilledema from normal and others (AUC 0.97 [0.95-0.99]; sensitivity 84.0% [75.0%-92.6%] and specificity 98.9% [98.5%-99.4%]). Six patients with missed papilledema in 1 eye were correctly identified by the deep learning system as having papilledema in the other eye. CONCLUSIONS: The BONSAI deep learning system was able to reliably identify papilledema and normal optic discs on nonmydriatic photographs obtained in the FOTO-ED studies. Our deep learning system has excellent potential as a diagnostic aid in EDs and non-ophthalmology clinics equipped with nonmydriatic fundus cameras.†.

Are isolated ocular motor nerve palsies in the elderly truly 'low risk' for abnormal neuro-imaging outcomes?

BACKGROUND: Neurologically isolated ocular motor nerve palsies often present a management dilemma. Neuroimaging is more likely to be offered to patients <50 years without coexisting ischaemic risk factors as their risk of sinister underlying causes is thought to be higher. However, populations are rapidly ageing and advanced neuroimaging is now more widely available. We thus investigated the incidence of abnormal neuroimaging outcomes in the traditionally low-risk older patient group. METHODS: This is a retrospective cohort study of 353 patients presenting with isolated ocular motor nerve palsies to a tertiary neuro-ophthalmology service in Singapore over a four-year (2015 to 2019) period. Clinical data was obtained through manual review of case records. Common aetiologies, age-based differences in prevalence of causes and abnormal neuroimaging outcomes were statistically analysed. RESULTS: Abnormal neuroimaging outcomes were significantly greater in the younger cohort only when age segregation was performed at 60 years of age. In a multivariate analysis, acute onset rather than ischaemic risk factors were independently predictive of normal neuroimaging outcomes. After adjusting for prior cancer risk and clinical bias from presumed ischaemic palsies, abnormal neuroimaging outcomes were seen in 14.1% ≥ 50 yrs, 10.9% ≥ 60 yrs and 15.1% ≥ 70 yrs. CONCLUSIONS: In patients presenting with isolated ocular motor nerve palsies, acute onset may be a more reliable indicator of an ischaemic palsy rather than advanced age or presence of ischaemic risk factors. If onset is not acute, neuroimaging should be considered irrespective of age and coexisting ischaemic risk factors.

Content development of the VISION-Cog: a novel tool to assess cognitive impairment in visually impaired older adults in Singapore.

OBJECTIVES: Current cognitive screening and diagnostic instruments rely on visually dependent tasks and are, therefore, not suitable to assess cognitive impairment (CI) in visually impaired older adults. We describe the content development of the VISually Independent test battery Of NeuroCOGnition (VISION-Cog)-a new diagnostic tool to evaluate CI in visually impaired older Singaporean adults. DESIGN: The content development phase consisted of two iterative stages: a neuropsychological consultation and literature review (stage 1) and an expert-panel discussion (stage 2). In stage 1, we investigated currently available neuropsychological test batteries for CI to inform constructions of our preliminary test battery. We then deliberated this battery during a consensus meeting using the Modified Nominal Group technique (stage 2) to decide, via agreement of five experts, the content of a pilot neuropsychological battery for the visually impaired. SETTING: Singapore Eye Research Institute. PARTICIPANTS: Stakeholders included researchers, psychologists, neurologists, neuro-ophthalmologists, geriatricians and psychiatrists. OUTCOME MEASURE: pilot VISION-Cog. RESULTS: The two-stage process resulted in a pilot VISION-Cog consisting of nine vision-independent neuropsychological tests, including the modified spatial memory test, list learning, list recall and list recognition, adapted token test, semantic fluency, modified spatial analysis, verbal subtests of the frontal battery assessment, digit symbol, digit span forwards, and digit span backwards. These tests encompassed five cognitive domains-memory and learning, language, executive function, complex attention, and perceptual-motor abilities. The expert panel suggested improvements to the clarity of test instructions and culturally relevant test content. These suggestions were incorporated and iteratively pilot-tested by the study team until no further issues emerged. CONCLUSIONS: We have developed a five-domain and nine-test VISION-Cog pilot instrument capable of replacing vision-dependent diagnostic batteries in aiding the clinician-based diagnosis of CI in visually impaired older adults. Subsequent phases will examine the VISION-Cog's feasibility, comprehensibility and acceptability; and evaluate its diagnostic performance.

AI-based diagnosis in mandibulofacial dysostosis with microcephaly using external ear shapes.

Introduction: Mandibulo-Facial Dysostosis with Microcephaly (MFDM) is a rare disease with a broad spectrum of symptoms, characterized by zygomatic and mandibular hypoplasia, microcephaly, and ear abnormalities. Here, we aimed at describing the external ear phenotype of MFDM patients, and train an Artificial Intelligence (AI)-based model to differentiate MFDM ears from non-syndromic control ears (binary classification), and from ears of the main differential diagnoses of this condition (multi-class classification): Treacher Collins (TC), Nager (NAFD) and CHARGE syndromes. Methods: The training set contained 1,592 ear photographs, corresponding to 550 patients. We extracted 48 patients completely independent of the training set, with only one photograph per ear per patient. After a CNN-(Convolutional Neural Network) based ear detection, the images were automatically landmarked. Generalized Procrustes Analysis was then performed, along with a dimension reduction using PCA (Principal Component Analysis). The principal components were used as inputs in an eXtreme Gradient Boosting (XGBoost) model, optimized using a 5-fold cross-validation. Finally, the model was tested on an independent validation set. Results: We trained the model on 1,592 ear photographs, corresponding to 1,296 control ears, 105 MFDM, 33 NAFD, 70 TC and 88 CHARGE syndrome ears. The model detected MFDM with an accuracy of 0.969 [0.838-0.999] (p < 0.001) and an AUC (Area Under the Curve) of 0.975 within controls (binary classification). Balanced accuracies were 0.811 [0.648-0.920] (p = 0.002) in a first multiclass design (MFDM vs. controls and differential diagnoses) and 0.813 [0.544-0.960] (p = 0.003) in a second multiclass design (MFDM vs. differential diagnoses). Conclusion: This is the first AI-based syndrome detection model in dysmorphology based on the external ear, opening promising clinical applications both for local care and referral, and for expert centers.

Feasibility, comprehensibility and acceptability of the VISION-Cog, a novel tool to assess cognitive impairment in visually impaired older adults: a cross-sectional pilot study in Singapore.

OBJECTIVES: We pilot-tested the VISually Independent test battery Of NeuroCOGnition (VISION-Cog) to determine its feasibility, comprehensibility and acceptability in evaluating cognitive impairment (CI) in visually impaired older Asian adults. DESIGN: The VISION-Cog was iteratively fine-tuned through pilot studies and expert-panel discussion. In the first pilot study (Stage 1), we recruited 15 visually impaired and cognitively normal participants aged ≥60 years to examine the pilot VISION-Cog's feasibility (length of time to administer), comprehensibility (clarity of instructions) and acceptability (participant burden). We then presented the pilot results to the expert panel (Stage 2) who decided via agreement on a revised version of the VISION-Cog. Subsequently, we conducted a second pilot study (Stage 3) on another four participants to ascertain improvement in feasibility, comprehensibility and acceptability of the revised version. SETTING: Singapore Eye Research Institute. PARTICIPANTS: Nineteen Asian adults aged ≥60 years with visual impairment (defined as near visual acuity worse than N8) were recruited. OUTCOME MEASURE: Revised VISION-Cog. RESULT: The VISION-Cog was deemed feasible, taking approximately 60 min to complete on average. All participants agreed that the test instructions were clear, and the battery did not cause undue discomfort or frustration. The data collector rated all tests as very user-friendly (score of 5/5). Minor modifications to the pilot VISION-Cog were suggested by the panel to improve its safety, clarity of instructions and content validity, which were incorporated and iteratively tested in the second pilot study until no further issues emerged. CONCLUSIONS: Using an iterative mixed-methods process, we have developed a feasible, comprehensible and acceptable 5-domain and 9-item visually independent VISION-Cog test battery suitable to assist CI diagnosis in older adults with visual impairment. We will assess its diagnostic potential against clinician-based assessment of CI in subsequent phases.

Handheld chromatic pupillometry can reliably detect functional glaucomatous damage in eyes with high myopia.

BACKGROUND/AIMS: To assess pupillary light responses (PLRs) in eyes with high myopia (HM) and evaluate the ability of handheld chromatic pupillometry (HCP) to identify glaucomatous functional loss in eyes with HM. METHODS: This prospective, cross-sectional study included 28 emmetropes (EM), 24 high myopes without glaucoma (HM) and 17 high myopes with confirmed glaucoma (HMG), recruited at the Singapore National Eye Center. Monocular PLRs were evaluated using a custom-built handheld pupillometer that recorded changes in horizontal pupil radius in response to 9 s of exponentially increasing blue (469.1 nm) and red (640.1 nm) lights. Fifteen pupillometric features were compared between groups. A logistic regression model (LRM) was used to distinguish HMG eyes from non-glaucomatous eyes (EM and HM). RESULTS: All pupillometric features were similar between EM and HM groups. Phasic constriction to blue (p<0.001) and red (p=0.006) lights, and maximum constriction to blue light (p<0.001) were reduced in HMG compared with EM and HM. Pupillometric features of melanopsin function (postillumination pupillary response, PIPR area under the curve (AUC) 0-12 s (p<0.001) and PIPR 6 s (p=0.01) to blue light) were reduced in HMG. Using only three pupillometric features, the LRM could classify glaucomatous from non-glaucomatous eyes with an AUC of 0.89 (95% CI 0.77 to 1.00), sensitivity 94.1% (95% CI 82.4% to 100.0%) and specificity 78.8% (95% CI 67.3% to 90.4%). CONCLUSION: PLRs to ramping-up light stimuli are unaltered in highly myopic eyes without other diagnosed ocular conditions. Conversely, HCP can distinguish glaucomatous functional loss in eyes with HM and can be a useful tool to detect/confirm the presence of glaucoma in patients with HM.

Adduction induces large optic nerve head deformations in subjects with normal-tension glaucoma.

PURPOSE: To assess intraocular pressure (IOP)-induced and gaze-induced optic nerve head (ONH) strains in subjects with high-tension glaucoma (HTG) and normal-tension glaucoma (NTG). DESIGN: Clinic-based cross-sectional study. METHODS: The ONH from one eye of 228 subjects (114 subjects with HTG (pre-treatment IOP≥21 mm Hg) and 114 with NTG (pre-treatment IOP<21 mm Hg)) was imaged with optical coherence tomography (OCT) under the following conditions: (1) OCT primary gaze, (2) 20° adduction from OCT primary gaze, (3) 20° abduction from OCT primary gaze and (4) OCT primary gaze with acute IOP elevation (to approximately 33 mm Hg). We then performed digital volume correlation analysis to quantify IOP-induced and gaze-induced ONH tissue deformations and strains. RESULTS: Across all subjects, adduction generated high effective strain (4.4%±2.3%) in the LC tissue with no significant difference (p>0.05) with those induced by IOP elevation (4.5%±2.4%); while abduction generated significantly lower (p=0.01) effective strain (3.1%±1.9%). The lamina cribrosa (LC) of HTG subjects exhibited significantly higher effective strain than those of NTG subjects under IOP elevation (HTG: 4.6%±1.7% vs NTG: 4.1%±1.5%, p<0.05). Conversely, the LC of NTG subjects exhibited significantly higher effective strain than those of HTG subjects under adduction (NTG: 4.9%±1.9% vs HTG: 4.0%±1.4%, p<0.05). CONCLUSION: We found that NTG subjects experienced higher strains due to adduction than HTG subjects, while HTG subjects experienced higher strain due to IOP elevation than NTG subjects-and that these differences were most pronounced in the LC tissue.

A Duration-Dependent Interaction Between High-Intensity Light and Unrestricted Vision in the Drive for Myopia Control.

Purpose: To evaluate the duration-dependent and synergetic impact of high-intensity light (HL) and unrestricted vision (UnV) on lens-induced myopia (LIM) development in chickens. Methods: Myopia was induced in one eye in chicks (10 groups, n = 126) from day 1 posthatching (D1) until day 8 (D8) using -10 diopter (D) lenses. Fellow eyes remained uncovered as controls. Nine groups were exposed daily to 2, 4, or 6 hours of HL (15,000 lux), UnV (removal of -10 D lens), or both (HL + UnV). One group served as the LIM group without any interventions. Ocular axial length (AL), refractive error, and choroidal thickness were measured on D1, D4, and D8. Outcome measures are expressed as interocular difference (IOD = experimental eye - control eye) ± SEM. Results: By D8, LIM increased AL (0.36 ± 0.04 mm), myopic refraction (-9.02 ± 0.37 D), and choroidal thinning (-90.27 ± 16.44 µm) in the LIM group (all, P < 0.001). Compared to the LIM group, exposure to 2, 4, or 6 hours of HL, UnV, or HL + UnV reduced myopic refraction in a duration-dependent manner, with UnV being more effective than HL (P < 0.05). Only 6 hours of HL + UnV (not 2 or 4 hours) prevented LIM and was more effective than UnV (P = 0.004) or HL (P < 0.001) in reducing myopic refraction and more effective than HL (P < 0.001) in reducing axial elongation. Conclusions: Daily exposure to 2, 4, or 6 hours of HL, UnV, or HL + UnV reduced lens-induced myopic refraction in a duration-dependent manner in chickens. Only 6 hours of HL + UnV completely stopped LIM development. The synergetic effect of HL and UnV is dependent on the duration of the interventions.

Through the eyes into the brain, using artificial intelligence.

INTRODUCTION: Detection of neurological conditions is of high importance in the current context of increasingly ageing populations. Imaging of the retina and the optic nerve head represents a unique opportunity to detect brain diseases, but requires specific human expertise. We review the current outcomes of artificial intelligence (AI) methods applied to retinal imaging for the detection of neurological and neuro-ophthalmic conditions. METHOD: Current and emerging concepts related to the detection of neurological conditions, using AI-based investigations of the retina in patients with brain disease were examined and summarised. RESULTS: Papilloedema due to intracranial hypertension can be accurately identified with deep learning on standard retinal imaging at a human expert level. Emerging studies suggest that patients with Alzheimer's disease can be discriminated from cognitively normal individuals, using AI applied to retinal images. CONCLUSION: Recent AI-based systems dedicated to scalable retinal imaging have opened new perspectives for the detection of brain conditions directly or indirectly affecting retinal structures. However, further validation and implementation studies are required to better understand their potential value in clinical practice.

A Deep Learning System for Automated Quality Evaluation of Optic Disc Photographs in Neuro-Ophthalmic Disorders.

The quality of ocular fundus photographs can affect the accuracy of the morphologic assessment of the optic nerve head (ONH), either by humans or by deep learning systems (DLS). In order to automatically identify ONH photographs of optimal quality, we have developed, trained, and tested a DLS, using an international, multicentre, multi-ethnic dataset of 5015 ocular fundus photographs from 31 centres in 20 countries participating to the Brain and Optic Nerve Study with Artificial Intelligence (BONSAI). The reference standard in image quality was established by three experts who independently classified photographs as of "good", "borderline", or "poor" quality. The DLS was trained on 4208 fundus photographs and tested on an independent external dataset of 807 photographs, using a multi-class model, evaluated with a one-vs-rest classification strategy. In the external-testing dataset, the DLS could identify with excellent performance "good" quality photographs (AUC = 0.93 (95% CI, 0.91-0.95), accuracy = 91.4% (95% CI, 90.0-92.9%), sensitivity = 93.8% (95% CI, 92.5-95.2%), specificity = 75.9% (95% CI, 69.7-82.1%) and "poor" quality photographs (AUC = 1.00 (95% CI, 0.99-1.00), accuracy = 99.1% (95% CI, 98.6-99.6%), sensitivity = 81.5% (95% CI, 70.6-93.8%), specificity = 99.7% (95% CI, 99.6-100.0%). "Borderline" quality images were also accurately classified (AUC = 0.90 (95% CI, 0.88-0.93), accuracy = 90.6% (95% CI, 89.1-92.2%), sensitivity = 65.4% (95% CI, 56.6-72.9%), specificity = 93.4% (95% CI, 92.1-94.8%). The overall accuracy to distinguish among the three classes was 90.6% (95% CI, 89.1-92.1%), suggesting that this DLS could select optimal quality fundus photographs in patients with neuro-ophthalmic and neurological disorders affecting the ONH.

Deep Learning System Outperforms Clinicians in Identifying Optic Disc Abnormalities.

BACKGROUND: The examination of the optic nerve head (optic disc) is mandatory in patients with headache, hypertension, or any neurological symptoms, yet it is rarely or poorly performed in general clinics. We recently developed a brain and optic nerve study with artificial intelligence-deep learning system (BONSAI-DLS) capable of accurately detecting optic disc abnormalities including papilledema (swelling due to elevated intracranial pressure) on digital fundus photographs with a comparable classification performance to expert neuro-ophthalmologists, but its performance compared to first-line clinicians remains unknown. METHODS: In this international, cross-sectional multicenter study, the DLS, trained on 14,341 fundus photographs, was tested on a retrospectively collected convenience sample of 800 photographs (400 normal optic discs, 201 papilledema and 199 other abnormalities) from 454 patients with a robust ground truth diagnosis provided by the referring expert neuro-ophthalmologists. The areas under the receiver-operating-characteristic curves were calculated for the BONSAI-DLS. Error rates, accuracy, sensitivity, and specificity of the algorithm were compared with those of 30 clinicians with or without ophthalmic training (6 general ophthalmologists, 6 optometrists, 6 neurologists, 6 internists, 6 emergency department [ED] physicians) who graded the same testing set of images. RESULTS: With an error rate of 15.3%, the DLS outperformed all clinicians (average error rates 24.4%, 24.8%, 38.2%, 44.8%, 47.9% for general ophthalmologists, optometrists, neurologists, internists and ED physicians, respectively) in the overall classification of optic disc appearance. The DLS displayed significantly higher accuracies than 100%, 86.7% and 93.3% of clinicians (n = 30) for the classification of papilledema, normal, and other disc abnormalities, respectively. CONCLUSIONS: The performance of the BONSAI-DLS to classify optic discs on fundus photographs was superior to that of clinicians with or without ophthalmic training. A trained DLS may offer valuable diagnostic aid to clinicians from various clinical settings for the screening of optic disc abnormalities harboring potentially sight- or life-threatening neurological conditions.

Handheld chromatic pupillometry can accurately and rapidly reveal functional loss in glaucoma.

BACKGROUND/AIMS: Early detection and treatment of glaucoma can delay vision loss. In this study, we evaluate the performance of handheld chromatic pupillometry (HCP) for the objective and rapid detection of functional loss in glaucoma. METHODS: In this clinic-based, prospective study, we enrolled 149 patients (median (IQR) years: 68.5 (13.6) years) with confirmed glaucoma and 173 healthy controls (55.2 (26.7) years). Changes in pupil size in response to 9 s of exponentially increasing blue (469 nm) and red (640 nm) light-stimuli were assessed monocularly using a custom-built handheld pupillometer. Pupillometric features were extracted from individual traces and compared between groups. Features with the highest classification potential, selected using a gradient boosting machine technique, were incorporated into a generalised linear model for glaucoma classification. Receiver operating characteristic curve analyses (ROC) were used to compare the performance of HCP, optical coherence tomography (OCT) and Humphrey Visual Field (HVF). RESULTS: Pupillary light responses were altered in glaucoma compared with controls. For glaucoma classification, HCP yielded an area under the ROC curve (AUC) of 0.94 (95% CI 0.91 to 0.96), a sensitivity of 87.9% and specificity of 88.4%. The classification performance of HCP in early-moderate glaucoma (visual field mean deviation (VFMD) > -12 dB; AUC=0.91 (95% CI 0.87 to 0.95)) was similar to HVF (AUC=0.91) and reduced compared with OCT (AUC=0.97; p=0.01). For severe glaucoma (VFMD ≤ -12 dB), HCP had an excellent classification performance (AUC=0.98, 95% CI 0.97 to 1) that was similar to HVF and OCT. CONCLUSION: HCP allows for an accurate, objective and rapid detection of functional loss in glaucomatous eyes of different severities.

The top 10 most frequently involved genes in hereditary optic neuropathies in 2186 probands.

Hereditary optic neuropathies are caused by the degeneration of retinal ganglion cells whose axons form the optic nerves, with a consistent genetic heterogeneity. As part of our diagnostic activity, we retrospectively evaluated the combination of Leber hereditary optic neuropathy mutations testing with the exon sequencing of 87 nuclear genes on 2186 patients referred for suspected hereditary optic neuropathies. The positive diagnosis rate in individuals referred for Leber hereditary optic neuropathy testing was 18% (199/1126 index cases), with 92% (184/199) carrying one of the three main pathogenic variants of mitochondrial DNA (m.11778G>A, 66.5%; m.3460G>A, 15% and m.14484T>C, 11%). The positive diagnosis rate in individuals referred for autosomal dominant or recessive optic neuropathies was 27% (451/1680 index cases), with 10 genes accounting together for 96% of this cohort. This represents an overall positive diagnostic rate of 30%. The identified top 10 nuclear genes included OPA1, WFS1, ACO2, SPG7, MFN2, AFG3L2, RTN4IP1, TMEM126A, NR2F1 and FDXR. Eleven additional genes, each accounting for less than 1% of cases, were identified in 17 individuals. Our results show that 10 major genes account for more than 96% of the cases diagnosed with our nuclear gene panel.

Discriminating Between Papilledema and Optic Disc Drusen Using 3D Structural Analysis of the Optic Nerve Head.

BACKGROUND AND OBJECTIVES: The distinction of papilledema from other optic nerve head (ONH) lesions mimicking papilledema, such as optic disc drusen (ODD), can be difficult in clinical practice. We aimed the following: (1) to develop a deep learning algorithm to automatically identify major structures of the ONH in 3-dimensional (3D) optical coherence tomography (OCT) scans and (2) to exploit such information to robustly differentiate among ODD, papilledema, and healthy ONHs. METHODS: This was a cross-sectional comparative study of patients from 3 sites (Singapore, Denmark, and Australia) with confirmed ODD, those with papilledema due to raised intracranial pressure, and healthy controls. Raster scans of the ONH were acquired using OCT imaging and then processed to improve deep-tissue visibility. First, a deep learning algorithm was developed to identify major ONH tissues and ODD regions. The performance of our algorithm was assessed using the Dice coefficient. Second, a classification algorithm (random forest) was designed to perform 3-class classifications (1: ODD, 2: papilledema, and 3: healthy ONHs) strictly from their drusen and prelamina swelling scores (calculated from the segmentations). To assess performance, we reported the area under the receiver operating characteristic curve for each class. RESULTS: A total of 241 patients (256 imaged ONHs, including 105 ODD, 51 papilledema, and 100 healthy ONHs) were retrospectively included in this study. Using OCT images of the ONH, our segmentation algorithm was able to isolate neural and connective tissues and ODD regions/conglomerates whenever present. This was confirmed by an averaged Dice coefficient of 0.93 ± 0.03 on the test set, corresponding to good segmentation performance. Classification was achieved with high AUCs, that is, 0.99 ± 0.001 for the detection of ODD, 0.99 ± 0.005 for the detection of papilledema, and 0.98 ± 0.01 for the detection of healthy ONHs. DISCUSSION: Our artificial intelligence approach can discriminate ODD from papilledema, strictly using a single OCT scan of the ONH. Our classification performance was very good in the studied population, with the caveat that validation in a much larger population is warranted. Our approach may have the potential to establish OCT imaging as one of the mainstays of diagnostic imaging for ONH disorders in neuro-ophthalmology, in addition to fundus photography.

Acceptance and Perception of Artificial Intelligence Usability in Eye Care (APPRAISE) for Ophthalmologists: A Multinational Perspective.

Background: Many artificial intelligence (AI) studies have focused on development of AI models, novel techniques, and reporting guidelines. However, little is understood about clinicians' perspectives of AI applications in medical fields including ophthalmology, particularly in light of recent regulatory guidelines. The aim for this study was to evaluate the perspectives of ophthalmologists regarding AI in 4 major eye conditions: diabetic retinopathy (DR), glaucoma, age-related macular degeneration (AMD) and cataract. Methods: This was a multi-national survey of ophthalmologists between March 1st, 2020 to February 29th, 2021 disseminated via the major global ophthalmology societies. The survey was designed based on microsystem, mesosystem and macrosystem questions, and the software as a medical device (SaMD) regulatory framework chaired by the Food and Drug Administration (FDA). Factors associated with AI adoption for ophthalmology analyzed with multivariable logistic regression random forest machine learning. Results: One thousand one hundred seventy-six ophthalmologists from 70 countries participated with a response rate ranging from 78.8 to 85.8% per question. Ophthalmologists were more willing to use AI as clinical assistive tools (88.1%, n = 890/1,010) especially those with over 20 years' experience (OR 3.70, 95% CI: 1.10-12.5, p = 0.035), as compared to clinical decision support tools (78.8%, n = 796/1,010) or diagnostic tools (64.5%, n = 651). A majority of Ophthalmologists felt that AI is most relevant to DR (78.2%), followed by glaucoma (70.7%), AMD (66.8%), and cataract (51.4%) detection. Many participants were confident their roles will not be replaced (68.2%, n = 632/927), and felt COVID-19 catalyzed willingness to adopt AI (80.9%, n = 750/927). Common barriers to implementation include medical liability from errors (72.5%, n = 672/927) whereas enablers include improving access (94.5%, n = 876/927). Machine learning modeling predicted acceptance from participant demographics with moderate to high accuracy, and area under the receiver operating curves of 0.63-0.83. Conclusion: Ophthalmologists are receptive to adopting AI as assistive tools for DR, glaucoma, and AMD. Furthermore, ML is a useful method that can be applied to evaluate predictive factors on clinical qualitative questionnaires. This study outlines actionable insights for future research and facilitation interventions to drive adoption and operationalization of AI tools for Ophthalmology.