Artificial intelligence for ocular oncology.

PURPOSE OF REVIEW: The aim of this article is to provide an update on the latest applications of deep learning (DL) and classical machine learning (ML) techniques to the detection and prognostication of intraocular and ocular surface malignancies. RECENT FINDINGS: Most recent studies focused on using DL and classical ML techniques for prognostication purposes in patients with uveal melanoma (UM). SUMMARY: DL has emerged as the leading ML technique for prognostication in ocular oncological conditions, particularly in UM. However, the application of DL may be limited by the relatively rarity of these conditions.

The Role of Ultra-Widefield Fundus Imaging and Fluorescein Angiography in Diagnosis and Treatment of Diabetic Retinopathy.

PURPOSE OF REVIEW: Early detection and treatment are important for preventing vision loss from diabetic retinopathy. Historically, the gold standard for grading diabetic retinopathy has been based on 7-field 30-degree color fundus photographs that capture roughly the central third of the retina. Our aim was to review recent literature on the role of ultra-widefield (allowing capture of up to 82% of the retina in one frame) fundus imaging in screening, prognostication, and treatment of diabetic retinopathy. RECENT FINDINGS: Ultra-widefield fundus imaging can capture peripheral retinal lesions outside the traditional 7-field photographs that may correlate with increased risk of diabetic retinopathy progression. The speed and ability to image through undilated pupils make ultra-widefield imaging attractive for tele-ophthalmology screening. Ultra-widefield fluorescein angiography may help guide targeted laser treatment in eyes with proliferative diabetic retinopathy. Ultra-widefield imaging has potential to help shape new diabetic retinopathy screening, staging, and treatment protocols.

The Impact of COVID-19 on Diabetic Retinopathy Monitoring and Treatment.

PURPOSE OF REVIEW: Diabetic retinopathy (DR) is one of the leading causes of vision loss worldwide. Although screening and early treatment guidelines for DR have significantly reduced the disease burden, restrictions related to the COVID-19 pandemic have changed real-world practice patterns in the management of DR. This review summarizes evolving guidelines and outcomes of the treatment of DR in the setting of the pandemic. RECENT FINDINGS: Intravitreal injections for DR have decreased significantly globally during the pandemic, ranging from approximately 30 to nearly 100% reduction, compared to corresponding timepoints in 2019. Most studies on functional outcomes show a decrease in visual acuity on delayed follow-up. Changing practice patterns in the management of DR has led to fewer intravitreal injections and overall reduction in visual acuity on follow-up. As COVID variants emerge, it will be necessary to continue evaluating practice guidelines.

Choriocapillaris flow loss in center-involving retinitis pigmentosa: a quantitative optical coherence tomography angiography study using a novel classification system.

PURPOSE: Choriocapillaris insufficiency may play a role in centripetal retinitis pigmentosa (RP) progression involving the fovea. However, the relationship between choriocapillaris integrity and foveal damage in RP is unclear. We examined the relationship between choriocapillaris flow and the presence of foveal photoreceptor involvement in RP. METHODS: We categorized the severity of central involvement in RP by the occurrence of foveal ellipsoid zone (EZ) disruption: present (severe RP) or absent (mild RP). Using optical coherence tomography angiography (OCTA, AngioVue, Optovue) in cases and unaffected age-matched controls, we compared vessel density (VD) between the groups using the generalized linear mixed model, controlling for age, gender, and scan quality. RESULTS: Fifty-seven eyes (20 severe RP, 18 mild RP, and 19 controls) were included. Foveal and parafoveal mean outer retinal thickness (µm) were lower in severe RP (fovea: 101.3 ± 14.5; parafovea: 68.4 ± 11.7) than controls (fovea: 161.2 ± 8.9; parafovea: 142.1 ± 11.8; p ≤ 0.001) and mild RP (fovea: 162.0 ± 14.7; parafovea: 116.8 ± 29.4; p ≤ 0.0001). Foveal choriocapillaris VD (%) was lower in severe RP (56.7 ± 6.8) than controls (69.9 ± 4.6; p = 0.008) and mild RP (65.3 ± 5.3; p = 0.01). The parafoveal choriocapillaris VD was lower in severe RP than controls (64.4 ± 5.9 vs. 68.3 ± 4.1; p = 0.04) but no different than in mild RP (p = 0.4). CONCLUSION: Choriocapillaris flow loss was associated with fovea-involving photoreceptor damage in RP. Further research is warranted to validate this putative association and clarify causation. Choriocapillaris imaging using OCTA may provide information to supplement structural OCT findings when evaluating subjects with RP in neuroprotective or regenerative clinical trials.

Detection of Optic Disc Abnormalities in Color Fundus Photographs Using Deep Learning.

BACKGROUND: To date, deep learning-based detection of optic disc abnormalities in color fundus photographs has mostly been limited to the field of glaucoma. However, many life-threatening systemic and neurological conditions can manifest as optic disc abnormalities. In this study, we aimed to extend the application of deep learning (DL) in optic disc analyses to detect a spectrum of nonglaucomatous optic neuropathies. METHODS: Using transfer learning, we trained a ResNet-152 deep convolutional neural network (DCNN) to distinguish between normal and abnormal optic discs in color fundus photographs (CFPs). Our training data set included 944 deidentified CFPs (abnormal 364; normal 580). Our testing data set included 151 deidentified CFPs (abnormal 71; normal 80). Both the training and testing data sets contained a wide range of optic disc abnormalities, including but not limited to ischemic optic neuropathy, atrophy, compressive optic neuropathy, hereditary optic neuropathy, hypoplasia, papilledema, and toxic optic neuropathy. The standard measures of performance (sensitivity, specificity, and area under the curve of the receiver operating characteristic curve (AUC-ROC)) were used for evaluation. RESULTS: During the 10-fold cross-validation test, our DCNN for distinguishing between normal and abnormal optic discs achieved the following mean performance: AUC-ROC 0.99 (95 CI: 0.98-0.99), sensitivity 94% (95 CI: 91%-97%), and specificity 96% (95 CI: 93%-99%). When evaluated against the external testing data set, our model achieved the following mean performance: AUC-ROC 0.87, sensitivity 90%, and specificity 69%. CONCLUSION: In summary, we have developed a deep learning algorithm that is capable of detecting a spectrum of optic disc abnormalities in color fundus photographs, with a focus on neuro-ophthalmological etiologies. As the next step, we plan to validate our algorithm prospectively as a focused screening tool in the emergency department, which if successful could be beneficial because current practice pattern and training predict a shortage of neuro-ophthalmologists and ophthalmologists in general in the near future.

Performance and Limitation of Machine Learning Algorithms for Diabetic Retinopathy Screening: Meta-analysis.

BACKGROUND: Diabetic retinopathy (DR), whose standard diagnosis is performed by human experts, has high prevalence and requires a more efficient screening method. Although machine learning (ML)-based automated DR diagnosis has gained attention due to recent approval of IDx-DR, performance of this tool has not been examined systematically, and the best ML technique for use in a real-world setting has not been discussed. OBJECTIVE: The aim of this study was to systematically examine the overall diagnostic accuracy of ML in diagnosing DR of different categories based on color fundus photographs and to determine the state-of-the-art ML approach. METHODS: Published studies in PubMed and EMBASE were searched from inception to June 2020. Studies were screened for relevant outcomes, publication types, and data sufficiency, and a total of 60 out of 2128 (2.82%) studies were retrieved after study selection. Extraction of data was performed by 2 authors according to PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses), and the quality assessment was performed according to the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2). Meta-analysis of diagnostic accuracy was pooled using a bivariate random effects model. The main outcomes included diagnostic accuracy, sensitivity, and specificity of ML in diagnosing DR based on color fundus photographs, as well as the performances of different major types of ML algorithms. RESULTS: The primary meta-analysis included 60 color fundus photograph studies (445,175 interpretations). Overall, ML demonstrated high accuracy in diagnosing DR of various categories, with a pooled area under the receiver operating characteristic (AUROC) ranging from 0.97 (95% CI 0.96-0.99) to 0.99 (95% CI 0.98-1.00). The performance of ML in detecting more-than-mild DR was robust (sensitivity 0.95; AUROC 0.97), and by subgroup analyses, we observed that robust performance of ML was not limited to benchmark data sets (sensitivity 0.92; AUROC 0.96) but could be generalized to images collected in clinical practice (sensitivity 0.97; AUROC 0.97). Neural network was the most widely used method, and the subgroup analysis revealed a pooled AUROC of 0.98 (95% CI 0.96-0.99) for studies that used neural networks to diagnose more-than-mild DR. CONCLUSIONS: This meta-analysis demonstrated high diagnostic accuracy of ML algorithms in detecting DR on color fundus photographs, suggesting that state-of-the-art, ML-based DR screening algorithms are likely ready for clinical applications. However, a significant portion of the earlier published studies had methodology flaws, such as the lack of external validation and presence of spectrum bias. The results of these studies should be interpreted with caution.

Controversies in artificial intelligence.

PURPOSE OF REVIEW: To review four recent controversial topics arising from deep learning applications in ophthalmology. RECENT FINDINGS: The controversies of four recent topics surrounding deep learning applications in ophthalmology are discussed, including the following: lack of explainability, limited generalizability, potential biases and protection of patient confidentiality in large-scale data transfer. SUMMARY: These controversial issues spanning the domains of clinical medicine, public health, computer science, ethics and legal issues, are complex and likely will benefit from an interdisciplinary approach if artificial intelligence in ophthalmology is to succeed over the next decade.

Deep Learning and Transfer Learning for Optic Disc Laterality Detection: Implications for Machine Learning in Neuro-Ophthalmology.

BACKGROUND: Deep learning (DL) has demonstrated human expert levels of performance for medical image classification in a wide array of medical fields, including ophthalmology. In this article, we present the results of our DL system designed to determine optic disc laterality, right eye vs left eye, in the presence of both normal and abnormal optic discs. METHODS: Using transfer learning, we modified the ResNet-152 deep convolutional neural network (DCNN), pretrained on ImageNet, to determine the optic disc laterality. After a 5-fold cross-validation, we generated receiver operating characteristic curves and corresponding area under the curve (AUC) values to evaluate performance. The data set consisted of 576 color fundus photographs (51% right and 49% left). Both 30° photographs centered on the optic disc (63%) and photographs with varying degree of optic disc centration and/or wider field of view (37%) were included. Both normal (27%) and abnormal (73%) optic discs were included. Various neuro-ophthalmological diseases were represented, such as, but not limited to, atrophy, anterior ischemic optic neuropathy, hypoplasia, and papilledema. RESULTS: Using 5-fold cross-validation (70% training; 10% validation; 20% testing), our DCNN for classifying right vs left optic disc achieved an average AUC of 0.999 (±0.002) with optimal threshold values, yielding an average accuracy of 98.78% (±1.52%), sensitivity of 98.60% (±1.72%), and specificity of 98.97% (±1.38%). When tested against a separate data set for external validation, our 5-fold cross-validation model achieved the following average performance: AUC 0.996 (±0.005), accuracy 97.2% (±2.0%), sensitivity 96.4% (±4.3%), and specificity 98.0% (±2.2%). CONCLUSIONS: Small data sets can be used to develop high-performing DL systems for semantic labeling of neuro-ophthalmology images, specifically in distinguishing between right and left optic discs, even in the presence of neuro-ophthalmological pathologies. Although this may seem like an elementary task, this study demonstrates the power of transfer learning and provides an example of a DCNN that can help curate large medical image databases for machine-learning purposes and facilitate ophthalmologist workflow by automatically labeling images according to laterality.

Ten-Year Outcome of Boston Type I Keratoprosthesis Surgery at a Tertiary Care Center.

PURPOSE: The aim of this study was to report long-term outcomes of patients who have undergone Boston type I keratoprosthesis (KPro) surgery. METHODS: This study was a retrospective review. Inclusion criteria were KPro surgery between 2006 and 2012 and at least 10 years of follow-up. Demographics, ocular history, surgery indication, clinical variables, and postsurgical outcomes were recorded. Descriptive statistical analysis was performed. RESULTS: We identified 75 patients with KPro implantation, and 17 patients with at least 10 years of follow-up (median = 11.1 years; range, 10.0-12.8 years) were included. Of 17 eyes, 11 (64.8%) had their original device in situ, 3 (17.6%) had their second device in situ, 1 (5.9%) had the device removed and replaced with a donor keratoplasty, and 2 (11.8%) were enucleated. At the last follow-up, 11 eyes (64.7%) were able to maintain improvement in vision, 5 (29.4%) had worsened vision, 1 (5.9%) had stable vision, and 9 (52.9%) had visual acuity

Retinal characteristics in eyes with pathologic myopia among individuals self-identifying as Black.

OBJECTIVE: Investigate retinal characteristics of pathologic myopia (PM) among patients self-identifying as Black. DESIGN: Retrospective cohort single-institution retrospective medical record review. METHODS: Adult patients between January 2005 and December 2014 with International Classification of Diseases (ICD) codes consistent with PM and given 5-year follow-up were evaluated. The Study Group consisted of patients self-identifying as Black, and the Comparison Group consisted of those not self-identifying as Black. Ocular features at study baseline and 5-year follow-up visit were evaluated. RESULTS: Among 428 patients with PM, 60 (14%) self-identified as Black and 18 (30%) had baseline and 5-year follow-up visits. Of the remaining 368 patients, 63 were in the Comparison Group. For the study (n = 18) and Comparison Group (n = 29), median (25th percentile, 75th percentile) baseline visual acuity was 20/40 (20/25, 20/50) and 20/32 (20/25, 20/50) in the better-seeing eye and 20/70 (20/50, 20/1400) and 20/100 (20/50, 20/200), respectively, in the worse-seeing eye. In the eyes that did not have choroidal neovascularization (CNV) in the study and Comparison Group, median study baseline optical coherence tomography central subfield thickness was 196 µm (169, 306 µm) and 225 µm (191, 280 µm), respectively, in the better-seeing eye and 208 µm (181, 260 µm) and 194 µm (171, 248 µm), respectively, in the worse-seeing eye. Baseline prevalence of CNV was 1 Study Group eye (3%) and 20 Comparison Group eyes (34%). By the 5-year visit, zero (0%) and 4 (15%) additional eyes had CNV in the study and Comparison Group, respectively. CONCLUSION: These findings suggest that the prevalence and incidence of CNV may be lower in patients with PM self-identifying as Black when compared with individuals of other races.

Autonomous Artificial Intelligence Increases Access and Health Equity in Underserved Populations with Diabetes.

Diabetic eye disease (DED) is a leading cause of blindness in the world. Early detection and treatment of DED have been shown to be both sight-saving and cost-effective. As such, annual testing for DED is recommended for adults with diabetes and is a Healthcare Effectiveness Data and Information Set (HEDIS) measure. However, adherence to this guideline has historically been low, and access to this sight-saving intervention has particularly been limited for specific populations, such as Black or African American patients. In 2018, the US Food and Drug Agency (FDA) De Novo cleared autonomous artificial intelligence (AI) for diagnosing DED in a primary care setting. In 2020, Johns Hopkins Medicine (JHM), an integrated healthcare system with over 30 primary care sites, began deploying autonomous AI for DED testing in some of its primary care clinics. In this retrospective study, we aimed to determine whether autonomous AI implementation was associated with increased adherence to annual DED testing, and whether this was different for specific populations. JHM primary care sites were categorized as "non-AI" sites (sites with no autonomous AI deployment over the study period and where patients are referred to eyecare for DED testing) or "AI-switched" sites (sites that did not have autonomous AI testing in 2019 but did by 2021). We conducted a difference-in-difference analysis using a logistic regression model to compare change in adherence rates from 2019 to 2021 between non-AI and AI-switched sites. Our study included all adult patients with diabetes managed within our health system (17,674 patients for the 2019 cohort and 17,590 patients for the 2021 cohort) and has three major findings. First, after controlling for a wide range of potential confounders, our regression analysis demonstrated that the odds ratio of adherence at AI-switched sites was 36% higher than that of non-AI sites, suggesting that there was a higher increase in DED testing between 2019 and 2021 at AI-switched sites than at non-AI sites. Second, our data suggested autonomous AI improved access for historically disadvantaged populations. The adherence rate for Black/African Americans increased by 11.9% within AI-switched sites whereas it decreased by 1.2% within non-AI sites over the same time frame. Third, the data suggest that autonomous AI improved health equity by closing care gaps. For example, in 2019, a large adherence rate gap existed between Asian Americans and Black/African Americans (61.1% vs. 45.5%). This 15.6% gap shrank to 3.5% by 2021. In summary, our real-world deployment results in a large integrated healthcare system suggest that autonomous AI improves adherence to a HEDIS measure, patient access, and health equity for patients with diabetes - particularly in historically disadvantaged patient groups. While our findings are encouraging, they will need to be replicated and validated in a prospective manner across more diverse settings.

Risk Factors for Nondiagnostic Imaging in a Real-World Deployment of Artificial Intelligence Diabetic Retinal Examinations in an Integrated Healthcare System: Maximizing Workflow Efficiency Through Predictive Dilation.

OBJECTIVE: In the pivotal clinical trial that led to Food and Drug Administration De Novo "approval" of the first fully autonomous artificial intelligence (AI) diabetic retinal disease diagnostic system, a reflexive dilation protocol was used. Using real-world deployment data before implementation of reflexive dilation, we identified factors associated with nondiagnostic results. These factors allow a novel predictive dilation workflow, where patients most likely to benefit from pharmacologic dilation are dilated a priori to maximize efficiency and patient satisfaction. METHODS: Retrospective review of patients who were assessed with autonomous AI at Johns Hopkins Medicine (8/2020 to 5/2021). We constructed a multivariable logistic regression model for nondiagnostic results to compare characteristics of patients with and without diagnostic results, using adjusted odds ratio (aOR). P < .05 was considered statistically significant. RESULTS: Of 241 patients (59% female; median age = 59), 123 (51%) had nondiagnostic results. In multivariable analysis, type 1 diabetes (T1D, aOR = 5.82, 95% confidence interval [CI]: 1.45-23.40, P = .01), smoking (aOR = 2.86, 95% CI: 1.36-5.99, P = .005), and age (every 10-year increase, aOR = 2.12, 95% CI: 1.62-2.77, P < .001) were associated with nondiagnostic results. Following feature elimination, a predictive model was created using T1D, smoking, age, race, sex, and hypertension as inputs. The model showed an area under the receiver-operator characteristics curve of 0.76 in five-fold cross-validation. CONCLUSIONS: We used factors associated with nondiagnostic results to design a novel, predictive dilation workflow, where patients most likely to benefit from pharmacologic dilation are dilated a priori. This new workflow has the potential to be more efficient than reflexive dilation, thus maximizing the number of at-risk patients receiving their diabetic retinal examinations.

Autonomous artificial intelligence increases screening and follow-up for diabetic retinopathy in youth: the ACCESS randomized control trial.

Diabetic retinopathy can be prevented with screening and early detection. We hypothesized that autonomous artificial intelligence (AI) diabetic eye exams at the point-of-care would increase diabetic eye exam completion rates in a racially and ethnically diverse youth population. AI for Children's diabetiC Eye ExamS (NCT05131451) is a parallel randomized controlled trial that randomized youth (ages 8-21 years) with type 1 and type 2 diabetes to intervention (autonomous artificial intelligence diabetic eye exam at the point of care), or control (scripted eye care provider referral and education) in an academic pediatric diabetes center. The primary outcome was diabetic eye exam completion rate within 6 months. The secondary outcome was the proportion of participants who completed follow-through with an eye care provider if deemed appropriate. Diabetic eye exam completion rate was significantly higher (100%, 95%CI: 95.5%, 100%) in the intervention group (n = 81) than the control group (n = 83) (22%, 95%CI: 14.2%, 32.4%)(p < 0.001). In the intervention arm, 25/81 participants had an abnormal result, of whom 64% (16/25) completed follow-through with an eye care provider, compared to 22% in the control arm (p < 0.001). Autonomous AI increases diabetic eye exam completion rates in youth with diabetes.

Use of Diabetes Technologies and Retinopathy in Adults With Type 1 Diabetes.

Importance: Diabetic retinopathy (DR) is a complication of diabetes that can lead to vision loss. Outcomes of continuous glucose monitoring (CGM) and insulin pump use in DR are not well understood. Objective: To assess the use of CGM, insulin pump, or both, and DR and proliferative diabetic retinopathy (PDR) in adults with type 1 diabetes (T1D). Design, Setting, and Participants: A retrospective cohort study of adults with T1D in a tertiary diabetes center and ophthalmology center was conducted from 2013 to 2021, with data analysis performed from June 2022 to April 2023. Exposure: Use of diabetes technologies, including insulin pump, CGM, and both CGM and insulin pump. Main Outcomes and Measures: The primary outcome was development of DR or PDR. A secondary outcome was the progression of DR for patients in the longitudinal cohort. Multivariable logistic regression models assessed for development of DR and PDR and association with CGM and insulin pump use. Results: A total of 550 adults with T1D were included (median age, 40 [IQR, 28-54] years; 54.4% female; 24.5% Black or African American; and 68.4% White), with a median duration of diabetes of 20 (IQR, 10-30) years, and median hemoglobin A1c (HbA1c) of 7.8% (IQR, 7.0%-8.9%). Overall, 62.7% patients used CGM, 58.2% used an insulin pump, and 47.5% used both; 44% (244 of 550) of the participants had DR at any point during the study. On univariate analysis, CGM use was associated with lower odds of DR and PDR, and CGM with pump was associated with lower odds of PDR (all P < .05), compared with no CGM use. Multivariable logistic regression adjusting for age, sex, race and ethnicity, diabetes duration, microvascular and macrovascular complications, insurance type, and mean HbA1c, showed that CGM was associated with lower odds of DR (odds ratio [OR], 0.52; 95% CI, 0.32-0.84; P = .008) and PDR (OR, 0.42; 95% CI, 0.23-0.75; P = .004), compared with no CGM use. In the longitudinal analysis of participants without baseline PDR, 79 of 363 patients (21.8%) had progression of DR during the study. Conclusions and Relevance: In this cohort study of adults with T1D, CGM use was associated with lower odds of developing DR and PDR, even after adjusting for HbA1c. These findings suggest that CGM may be useful for diabetes management to mitigate risk for DR and PDR.

Deep Learning Applications to Classification and Detection of Age-Related Macular Degeneration on Optical Coherence Tomography Imaging: A Review.

Age-related macular degeneration (AMD) is one of the leading causes of blindness in the elderly, more commonly in developed countries. Optical coherence tomography (OCT) is a non-invasive imaging device widely used for the diagnosis and management of AMD. Deep learning (DL) uses multilayered artificial neural networks (NN) for feature extraction, and is the cutting-edge technique for medical image analysis for diagnostic and prognostication purposes. Application of DL models to OCT image analysis has garnered significant interest in recent years. In this review, we aimed to summarize studies focusing on DL models used in classification and detection of AMD. Additionally, we provide a brief introduction to other DL applications in AMD, such as segmentation, prediction/prognostication, and models trained on multimodal imaging.

Correlation between hyperreflective foci and visual function testing in eyes with intermediate age-related macular degeneration.

BACKGROUND: To investigate the relationship between intraretinal hyperreflective foci (HRF) and visual function in intermediate age-related macular degeneration (iAMD). METHODS: Retrospective, cross-sectional study. iAMD patients underwent spectral domain optical coherence tomography (SD-OCT) imaging and vision function testing: normal luminance best corrected visual acuity (VA), low luminance VA (LLVA), quantitative contrast sensitivity function (qCSF), low luminance qCSF (LLqCSF), and mesopic microperimetry. Each OCT volume was graded for the presence and number of HRF. Each HRF was graded for: separation from the retinal pigment epithelium (RPE), above drusen, and shadowing. Central drusen volume was calculated by the built-in functionality of the commercial OCT software after manual segmentation of the RPE and Bruch's membrane. RESULTS: HRF group: 11 eyes; 9 patients; mean age 75.7 years. No-HRF group: 11 eyes; 10 patients; mean age 74.8 years. In linear mixed effect model adjusting for cube-root transformed drusen volume, HRF group showed statistically significant worse VA, LLVA, LLqCSF, and microperimetry. HRF group showed worse cone function, as measured by our pre-defined multicomponent endpoint, incorporating LLVA, LLqCSF and microperimetry (p = 0.018). For eyes with HRF, # of HRF did not correlate with any functional measures; however, % of HRF separated from RPE and # of HRF that created shadowing were statistically associated with low luminance deficit (LLD). CONCLUSIONS: The association between the presence of HRF and worse cone visual function supports the hypothesis that eyes with HRF have more advanced disease.

Vision transformers: The next frontier for deep learning-based ophthalmic image analysis.

Deep learning is the state-of-the-art machine learning technique for ophthalmic image analysis, and convolutional neural networks (CNNs) are the most commonly utilized approach. Recently, vision transformers (ViTs) have emerged as a promising approach, one that is even more powerful than CNNs. In this focused review, we summarized studies that applied ViT-based models to analyze color fundus photographs and optical coherence tomography images. Overall, ViT-based models showed robust performances in the grading of diabetic retinopathy and glaucoma detection. While some studies demonstrated that ViTs were superior to CNNs in certain contexts of use, it is unclear how widespread ViTs will be adopted for ophthalmic image analysis, since ViTs typically require even more training data as compared to CNNs. The studies included were identified from the PubMed and Google Scholar databases using keywords relevant to this review. Only original investigations through March 2023 were included.

Direct Gene Expression Profile Prediction for Uveal Melanoma from Digital Cytopathology Images via Deep Learning and Salient Image Region Identification.

Objective: To demonstrate that deep learning (DL) methods can produce robust prediction of gene expression profile (GEP) in uveal melanoma (UM) based on digital cytopathology images. Design: Evaluation of a diagnostic test or technology. Subjects Participants and Controls: Deidentified smeared cytology slides stained with hematoxylin and eosin obtained from a fine needle aspirated from UM. Methods: Digital whole-slide images were generated by fine-needle aspiration biopsies of UM tumors that underwent GEP testing. A multistage DL system was developed with automatic region-of-interest (ROI) extraction from digital cytopathology images, an attention-based neural network, ROI feature aggregation, and slide-level data augmentation. Main Outcome Measures: The ability of our DL system in predicting GEP on a slide (patient) level. Data were partitioned at the patient level (73% training; 27% testing). Results: In total, our study included 89 whole-slide images from 82 patients and 121 388 unique ROIs. The testing set included 24 slides from 24 patients (12 class 1 tumors; 12 class 2 tumors; 1 slide per patient). Our DL system for GEP prediction achieved an area under the receiver operating characteristic curve of 0.944, an accuracy of 91.7%, a sensitivity of 91.7%, and a specificity of 91.7% on a slide-level analysis. The incorporation of slide-level feature aggregation and data augmentation produced a more predictive DL model (P = 0.0031). Conclusions: Our current work established a complete pipeline for GEP prediction in UM tumors: from automatic ROI extraction from digital cytopathology whole-slide images to slide-level predictions. Our DL system demonstrated robust performance and, if validated prospectively, could serve as an image-based alternative to GEP testing.

Swept Source Optical Coherence Tomography Angiography Findings in Birdshot Chorioretinitis: A Cross Sectional Study of 21 Patients.

PURPOSE: Birdshot chorioretinitis (BSCR) is a form of posterior uveitis that is classically characterized by hypopigmented choroidal lesions outside of the major arcades. However, little is known about the extent of choroidal involvement in the macula. We aim to describe the vascular abnormalities observed at the level of the choriocapillaris (CC) in the maculae of BSCR patients, using swept source optical coherence tomography angiography (SS-OCTA). METHODS: A cross-sectional, observational study was conducted. Eligible patients underwent clinical examination and SS-OCTA imaging. The main outcome measures were the frequency of vascular abnormalities observed at the level of the CC on SS-OCTA and foveal choriocapillaris vascular density (CVD). RESULTS: Twenty-one patients were included, with a median age of 61.5 years. All patients had bilateral disease with a median disease duration of 6 years. All but one patient received systemic immunosuppressive drug therapy, and 19 patients had suppressed inflammation on treatment at the time of the SS-OCTA assessment. Of the 42 affected eyes, 39 (92.9%) had gradable SS-OCTA images, with a mean LogMAR visual acuity of 0.18 (Snellen equivalent 20/30). In total, 34 of 39 (87.2%) eyes had some degree of pathologic flow loss, and after controlling for patient age and disease activity, both worse VA and longer disease duration remained statistically significantly associated with reduced foveal CVD. CONCLUSIONS: Our findings suggest that pathologic CC flow loss in the macula is frequently encountered and may contribute to visual function decline in patients with BSCR. Further studies with longitudinal follow-up are needed to characterize the evolution of these areas of pathologic CC flow loss over time.