[Challenges and prospects in the application of artificial intelligence for ocular disease screening and diagnosis].

In recent years, artificial intelligence (AI) technologies have experienced substantial growth across various sectors, with significant strides made particularly in medical AI through advancements such as large models. The application of AI within the field of ophthalmology can enhance the accuracy of eye disease screening and diagnosis. However, the deployment of AI and its large models in ophthalmology still encounters numerous limitations and challenges. This article builds upon the transformative achievements in the medical AI sector and discusses the ongoing challenges faced by AI applications in ophthalmology. It provides forward-looking insights from an ophthalmic perspective regarding the era of large models and anticipates research trends in AI applications in ophthalmology, so as to foster the continuous advancement of AI technologies, thereby significantly promoting eye health.

Raman spectroscopy assisted tear analysis: A label free, optical approach for noninvasive disease diagnostics.

In recent times, tear fluid analysis has garnered considerable attention in the field of biomarker-based diagnostics due to its noninvasive sample collection method. Tears encompass a reservoir of biomarkers that assist in diagnosing not only ocular disorders but also a diverse list of systemic diseases. This highlights the necessity for sensitive and dependable screening methods to employ tear fluid as a potential noninvasive diagnostic specimen in clinical environments. Considerable research has been conducted to investigate the potential of Raman spectroscopy-based investigations for tear analysis in various diagnostic applications. Raman Spectroscopy (RS) is a highly sensitive and label free spectroscopic technique which aids in investigating the molecular structure of samples by evaluating the vibrational frequencies of molecular bonds. Due to the distinct chemical compositions of different samples, it is possible to obtain a sample-specific spectral fingerprint. The distinctive spectral fingerprints obtained from Raman spectroscopy enable researchers to identify specific compounds or functional groups present in a sample, aiding in diverse biomedical applications. Its sensitivity to changes in molecular structure or environment provides invaluable insights into subtle alterations associated with various diseases. Thus, Raman Spectroscopy has the potential to assist in diagnosis and treatment as well as prognostic evaluation. Raman spectroscopy possesses several advantages, such as the non-destructive examination of samples, remarkable sensitivity to structural variations, minimal prerequisites for sample preparation, negligible interference from water, and the aptness for real-time investigation of tear samples. The purpose of this review is to highlight the potential of Raman spectroscopic technique in facilitating the clinical diagnosis of various ophthalmic and systemic disorders through non-invasive tear analysis. Additionally, the review delves into the advancements made in Raman spectroscopy with regards to paper-based sensing substrates and tear analysis methods integrated into contact lenses. Furthermore, the review also addresses the obstacles and future possibilities associated with implementing Raman spectroscopy as a routine diagnostic tool based on tear analysis in clinical settings.

New disposable esthesiometer (KeraSenseⓇ) to improve diagnosis and management of neurotrophic keratitis.

PURPOSE: To validate the use, repeatability, and reproducibility of a new, cost-effective, disposable, sterile device (KeraSenseⓇ, Dompè farmaceutici SpA, Milan Italy) compared to Cochet-Bonnet (CB) esthesiometer. Secondly, to identify a simple, safe, rapid, and low-cost test to diagnose neurotrophic keratitis (NK). METHODS: 16 patients with diagnosis of NK stage I, 25 patients with diabetes mellitus (DM), and 26 healthy subjects were included in the study. Corneal sensitivity (CS) was assessed by CB and KeraSenseⓇ. Repeatability, accuracy, and reproducibility of the novel disposable aesthesiometer were assessed. Specificity, sensitivity, and cut-off value for NK diagnosis were calculated by ROC curve analysis. RESULTS: All NK patients showed a CS ≤ 40 mm, while none of the healthy patients showed a CS value < 50 mm. Significant agreement was found between CB measurements and the single use esthesiometer evaluations of CS (p < 0.001). Repeatability evaluations of the single use esthesiometer showed 100% agreement between different measurements (p < 0.001). Reproducibility evaluations showed 99.6% concordance between different operators (p < 0.001). A 55 mm value of the single use esthesiometer was adequate to exclude an NK diagnosis, while all NK patients showed a value ≤ 35 mm. CONCLUSIONS: Corneal hypo/anaesthesia is considered the hallmark of NK. The use of the novel single-use esthesiometer will allow for a diagnostic improvement in NK, sparing time and guaranteeing patients' safety. Diabetic patients despite normal corneal findings may show impairment of CS, suggesting a preclinical stage of NK, requiring a close follow-up.

Insights into the diagnosis and management of sarcoid uveitis: A review.

Sarcoidosis is a leading cause of non-infectious uveitis that commonly affects middle-aged individuals and has a female preponderance. The disease demonstrates age, sex and ethnic differences in clinical manifestations. A diagnosis of sarcoidosis is made based on a compatible clinical presentation, supporting investigations and histologic evidence of non-caseating granulomas, although biopsy is not always possible. Multimodal imaging with widefield fundus photography, optical coherence tomography and angiography can help in the diagnosis of sarcoid uveitis and in the monitoring of treatment response. Corticosteroid remains the mainstay of treatment; chronic inflammation requires steroid-sparing immunosuppression. Features on multimodal imaging such as vascular leakage may provide prognostic indicators of outcome. Female gender, prolonged and severe uveitis, and posterior involving uveitis are associated with poorer visual outcomes.

High Dynamic Range Image Processing for Retinal Color Fundus Photography.

BACKGROUND AND OBJECTIVE: Color fundus photography is an important imaging modality that is currently limited by a narrow dynamic range. We describe a post-image processing technique to generate high dynamic range (HDR) retinal images with enhanced detail. PATIENTS AND METHODS: This was a retrospective, observational case series evaluating fundus photographs of patients with macular pathology. Photographs were acquired with three or more exposure values using a commercially available camera (Topcon 50-DX). Images were aligned and imported into HDR processing software (Photomatix Pro). Fundus detail was compared between HDR and raw photographs. RESULTS: Sixteen eyes from 10 patients (5 male, 5 female; mean age 59.4 years) were analyzed. Clinician graders preferred the HDR image 91.7% of the time (44/48 image comparisons), with good grader agreement (81.3%, 13/16 eyes). CONCLUSIONS: HDR fundus imaging is feasible using images from existing fundus cameras and may be useful for enhanced visualization of retinal detail in a variety of pathologic states. [Ophthalmic Surg Lasers Imaging Retina 2024;55:263-269.].

Pupil size measurements with a multifunctional aberrometer/coherence interferometer/tomographer and two infrared-based pupillometers.

PURPOSE: To compare precision of pupil size measurements of a multifunctional device (Pentacam AXL Wave [Pentacam]) and 2 infrared-based pupillometers (PupilX, Colvard) and to compare repeatability of Pentacam and PupilX. SETTING: Department of Ophthalmology, Goethe-University, Frankfurt am Main, Germany. DESIGN: Prospective, comparative trial. METHODS: Pupil diameter of healthy eyes was measured with Colvard once and Pentacam without glare (WO) and with glare (WG), PupilX in 0, 1, and 16 lux 3 times each. In a second series, measurements with Pentacam WO and PupilX in 0.06 and 0.12 lux were assessed. RESULTS: 36 eyes of participants aged 21 to 63 years were included. Mean pupil diameter was 6.05 mm with Colvard, 5.79 mm (first series), 5.50 mm (second series) with Pentacam WO, 3.42 mm WG, 7.26 mm PupilX in 0, 4.67 mm 1, 3.66 mm 16, 6.82 mm in 0.06, and 6.39 mm in 0.12 lux. Measurements with Pentacam WO were significantly different to PupilX in 0, 0.06, 0.12, and 1 lux (all P < .001), but not to Colvard ( P = .086). Pupil size measured with Pentacam WG and PupilX in 16 lux was not significantly different ( P = .647). Consecutive measurements with Pentacam WO and WG had mean SD of 0.23 mm and 0.20 mm, respectively, and with PupilX 0.11 in 0, 0.24 mm 1, and 0.20 mm in 16 lux. CONCLUSIONS: Pentacam provided good assessment of pupil size but was not equivalent to PupilX in low lighting conditions. Repeatability was more favorable for Pentacam.

SSiT: Saliency-Guided Self-Supervised Image Transformer for Diabetic Retinopathy Grading.

Self-supervised Learning (SSL) has been widely applied to learn image representations through exploiting unlabeled images. However, it has not been fully explored in the medical image analysis field. In this work, Saliency-guided Self-Supervised image Transformer (SSiT) is proposed for Diabetic Retinopathy (DR) grading from fundus images. We novelly introduce saliency maps into SSL, with a goal of guiding self-supervised pre-training with domain-specific prior knowledge. Specifically, two saliency-guided learning tasks are employed in SSiT: 1) Saliency-guided contrastive learning is conducted based on the momentum contrast, wherein fundus images' saliency maps are utilized to remove trivial patches from the input sequences of the momentum-updated key encoder. Thus, the key encoder is constrained to provide target representations focusing on salient regions, guiding the query encoder to capture salient features. 2) The query encoder is trained to predict the saliency segmentation, encouraging the preservation of fine-grained information in the learned representations. To assess our proposed method, four publicly-accessible fundus image datasets are adopted. One dataset is employed for pre-training, while the three others are used to evaluate the pre-trained models' performance on downstream DR grading. The proposed SSiT significantly outperforms other representative state-of-the-art SSL methods on all downstream datasets and under various evaluation settings. For example, SSiT achieves a Kappa score of 81.88% on the DDR dataset under fine-tuning evaluation, outperforming all other ViT-based SSL methods by at least 9.48%.

Implementation of optical coherence tomography in retinopathy of prematurity screening.

PURPOSE OF REVIEW: In this review, we explore the investigational applications of optical coherence tomography (OCT) in retinopathy of prematurity (ROP), the insights they have delivered thus far, and key milestones for its integration into the standard of care. RECENT FINDINGS: While OCT has been widely integrated into clinical management of common retinal diseases, its use in pediatric contexts has been undermined by limitations in ergonomics, image acquisition time, and field of view. Recently, investigational handheld OCT devices have been reported with advancements including ultra-widefield view, noncontact use, and high-speed image capture permitting real-time en face visualization. These developments are compelling for OCT as a more objective alternative with reduced neonatal stress compared to indirect ophthalmoscopy and/or fundus photography as a means of classifying and monitoring ROP. SUMMARY: OCT may become a viable modality in management of ROP. Ongoing innovation surrounding handheld devices should aim to optimize patient comfort and image resolution in the retinal periphery. Future clinical investigations may seek to objectively characterize features of peripheral stage and explore novel biomarkers of disease activity.

Geometric Correspondence-Based Multimodal Learning for Ophthalmic Image Analysis.

Color fundus photography (CFP) and Optical coherence tomography (OCT) images are two of the most widely used modalities in the clinical diagnosis and management of retinal diseases. Despite the widespread use of multimodal imaging in clinical practice, few methods for automated diagnosis of eye diseases utilize correlated and complementary information from multiple modalities effectively. This paper explores how to leverage the information from CFP and OCT images to improve the automated diagnosis of retinal diseases. We propose a novel multimodal learning method, named geometric correspondence-based multimodal learning network (GeCoM-Net), to achieve the fusion of CFP and OCT images. Specifically, inspired by clinical observations, we consider the geometric correspondence between the OCT slice and the CFP region to learn the correlated features of the two modalities for robust fusion. Furthermore, we design a new feature selection strategy to extract discriminative OCT representations by automatically selecting the important feature maps from OCT slices. Unlike the existing multimodal learning methods, GeCoM-Net is the first method that formulates the geometric relationships between the OCT slice and the corresponding region of the CFP image explicitly for CFP and OCT fusion. Experiments have been conducted on a large-scale private dataset and a publicly available dataset to evaluate the effectiveness of GeCoM-Net for diagnosing diabetic macular edema (DME), impaired visual acuity (VA) and glaucoma. The empirical results show that our method outperforms the current state-of-the-art multimodal learning methods by improving the AUROC score 0.4%, 1.9% and 2.9% for DME, VA and glaucoma detection, respectively.

Diagnostic methods for primary vitreoretinal lymphoma: A systematic review.

Primary vitreoretinal lymphoma is a potentially aggressive intraocular malignancy with poor systemic prognosis and sometimes significant diagnostic delays as it may masquerade as chronic uveitis. Despite the variety of diagnostic techniques, it is unclear which modality is most accurate in the diagnosis of PVRL. A systematic literature search was conducted on Ovid MEDLINE, EMBASE and the Cochrane Controlled Register of Trials for studies published between January, 2000, and June, 2023. Randomized controlled trials (RCTs) reporting on the following diagnostic tools used to diagnose patients with PVRL were included: cytology, flow cytometry, MYD88 L265P mutation, CD79B mutation, interleukin 10/interleukin-6 (IL-10/IL-6) ratio, polymerase chain reaction (PCR) for monoclonal immunoglobulin heavy chain (IgH) and immunoglobulin kappa light chain (IgK) rearrangements, and imaging findings. The aggregated sensitivity of each diagnostic modality was reported and compared using the chi-squared (χ2) test. A total of 662 eyes from 29 retrospective studies reporting on patients diagnosed with PVRL were included. An IL-10/IL-6 ratio greater than 1 had the highest sensitivity (89.39%, n = 278/311 eyes, n = 16 studies) for PVRL, where the sensitivity was not significantly different when only vitreous samples were drawn (88.89%, n = 232/261 eyes, n = 13 studies) compared to aqueous samples (83.33%, n = 20/24, n = 2) (p = 0.42). Flow cytometry of vitreous samples gave a positive result in 66/75 eyes (88.00%, n = 6 studies) with PVRL, and monoclonal IgH rearrangements on PCR gave a positive result in 354/416 eyes (85.10%, n = 20 studies) with PVRL. MYD88 L265P and CD79B mutation analysis performed poorly, yielding a positive result in 63/90 eyes (70.00%, n = 8 studies) with PVRL, and 20/57 eyes (35.09%, n = 4 studies) with PVRL, respectively. Overall, our systematic review found that an IL-10/IL-6 ratio greater or equal to one may provide the highest sensitivity in identifying patients with PVRL. Future studies are needed to employ multiple diagnostic tools to aid in the detection of PVRL and to further establish nuanced guidelines when determining the optimal diagnostic tool to use in diverse patient populations.

Correction Method for Optical Scaling of Fundoscopy Images: Development, Validation, and First Implementation.

Purpose: Although fundus photography is extensively used in ophthalmology, refraction prevents accurate distance measurement on fundus images, as the resulting scaling differs between subjects due to varying ocular anatomy. We propose a PARaxial Optical fundus Scaling (PAROS) method to correct for this variation using commonly available clinical data. Methods: The complete optics of the eye and fundus camera were modeled using ray transfer matrix formalism to obtain fundus image magnification. The subject's ocular geometry was personalized using biometry, spherical equivalent of refraction (RSE), keratometry, and/or corneal topography data. The PAROS method was validated using 41 different eye phantoms and subsequently evaluated in 44 healthy phakic subjects (of whom 11 had phakic intraocular lenses [pIOLs]), 29 pseudophakic subjects, and 21 patients with uveal melanoma. Results: Validation of the PAROS method showed small differences between model and actual image magnification (maximum 3.3%). Relative to the average eye, large differences in fundus magnification were observed, ranging from 0.79 to 1.48. Magnification was strongly inversely related to RSE (R2 = 0.67). In phakic subjects, magnification was directly proportional to axial length (R2 = 0.34). The inverse relation was seen in pIOL (R2 = 0.79) and pseudophakic (R2 = 0.12) subjects. RSE was a strong contributor to magnification differences (1%-83%). As this effect is not considered in the commonly used Bennett-Littmann method, statistically significant differences up to 40% (mean absolute 9%) were observed compared to the PAROS method (P < 0.001). Conclusions: The significant differences in fundus image scaling observed among subjects can be accurately accounted for with the PAROS method, enabling more accurate quantitative assessment of fundus photography.

Comparison of quantification of intraretinal hard exudates between optical coherence tomography en face image versus fundus photography.

PURPOSE: To compare the quantification of intraretinal hard exudate (HE) using en face optical coherence tomography (OCT) and fundus photography. METHODS: Consecutive en face images and corresponding fundus photographs from 13 eyes of 10 patients with macular edema associated with diabetic retinopathy or Coats' disease were analyzed using the machine-learning-based image analysis tool, "ilastik." RESULTS: The overall measured HE area was greater with en face images than with fundus photos (en face: 0.49 ± 0.35 mm2 vs. fundus photo: 0.34 ± 0.34 mm2, P < 0.001). However, there was an excellent correlation between the two measurements (intraclass correlation coefficient [ICC] = 0.844). There was a negative correlation between HE area and central macular thickness (CMT) (r = -0.292, P = 0.001). However, HE area showed a positive correlation with CMT in the previous several months, especially in eyes treated with anti-vascular endothelial growth factor (VEGF) therapy (CMT 3 months before: r = 0.349, P = 0.001; CMT 4 months before: r = 0.287, P = 0.012). CONCLUSION: Intraretinal HE can be reliably quantified from either en face OCT images or fundus photography with the aid of an interactive machine learning-based image analysis tool. HE area changes lagged several months behind CMT changes, especially in eyes treated with anti-VEGF injections.

Development of a generative deep learning model to improve epiretinal membrane detection in fundus photography.

BACKGROUND: The epiretinal membrane (ERM) is a common retinal disorder characterized by abnormal fibrocellular tissue at the vitreomacular interface. Most patients with ERM are asymptomatic at early stages. Therefore, screening for ERM will become increasingly important. Despite the high prevalence of ERM, few deep learning studies have investigated ERM detection in the color fundus photography (CFP) domain. In this study, we built a generative model to enhance ERM detection performance in the CFP. METHODS: This deep learning study retrospectively collected 302 ERM and 1,250 healthy CFP data points from a healthcare center. The generative model using StyleGAN2 was trained using single-center data. EfficientNetB0 with StyleGAN2-based augmentation was validated using independent internal single-center data and external datasets. We randomly assigned healthcare center data to the development (80%) and internal validation (20%) datasets. Data from two publicly accessible sources were used as external validation datasets. RESULTS: StyleGAN2 facilitated realistic CFP synthesis with the characteristic cellophane reflex features of the ERM. The proposed method with StyleGAN2-based augmentation outperformed the typical transfer learning without a generative adversarial network. The proposed model achieved an area under the receiver operating characteristic (AUC) curve of 0.926 for internal validation. AUCs of 0.951 and 0.914 were obtained for the two external validation datasets. Compared with the deep learning model without augmentation, StyleGAN2-based augmentation improved the detection performance and contributed to the focus on the location of the ERM. CONCLUSIONS: We proposed an ERM detection model by synthesizing realistic CFP images with the pathological features of ERM through generative deep learning. We believe that our deep learning framework will help achieve a more accurate detection of ERM in a limited data setting.

Deep Learning Identifies High-Quality Fundus Photographs and Increases Accuracy in Automated Primary Open Angle Glaucoma Detection.

Purpose: To develop and evaluate a deep learning (DL) model to assess fundus photograph quality, and quantitatively measure its impact on automated POAG detection in independent study populations. Methods: Image quality ground truth was determined by manual review of 2815 fundus photographs of healthy and POAG eyes from the Diagnostic Innovations in Glaucoma Study and African Descent and Glaucoma Evaluation Study (DIGS/ADAGES), as well as 11,350 from the Ocular Hypertension Treatment Study (OHTS). Human experts assessed a photograph as high quality if of sufficient quality to determine POAG status and poor quality if not. A DL quality model was trained on photographs from DIGS/ADAGES and tested on OHTS. The effect of DL quality assessment on DL POAG detection was measured using area under the receiver operating characteristic (AUROC). Results: The DL quality model yielded an AUROC of 0.97 for differentiating between high- and low-quality photographs; qualitative human review affirmed high model performance. Diagnostic accuracy of the DL POAG model was significantly greater (P < 0.001) in good (AUROC, 0.87; 95% CI, 0.80-0.92) compared with poor quality photographs (AUROC, 0.77; 95% CI, 0.67-0.88). Conclusions: The DL quality model was able to accurately assess fundus photograph quality. Using automated quality assessment to filter out low-quality photographs increased the accuracy of a DL POAG detection model. Translational Relevance: Incorporating DL quality assessment into automated review of fundus photographs can help to decrease the burden of manual review and improve accuracy for automated DL POAG detection.

AIROGS: Artificial Intelligence for Robust Glaucoma Screening Challenge.

The early detection of glaucoma is essential in preventing visual impairment. Artificial intelligence (AI) can be used to analyze color fundus photographs (CFPs) in a cost-effective manner, making glaucoma screening more accessible. While AI models for glaucoma screening from CFPs have shown promising results in laboratory settings, their performance decreases significantly in real-world scenarios due to the presence of out-of-distribution and low-quality images. To address this issue, we propose the Artificial Intelligence for Robust Glaucoma Screening (AIROGS) challenge. This challenge includes a large dataset of around 113,000 images from about 60,000 patients and 500 different screening centers, and encourages the development of algorithms that are robust to ungradable and unexpected input data. We evaluated solutions from 14 teams in this paper and found that the best teams performed similarly to a set of 20 expert ophthalmologists and optometrists. The highest-scoring team achieved an area under the receiver operating characteristic curve of 0.99 (95% CI: 0.98-0.99) for detecting ungradable images on-the-fly. Additionally, many of the algorithms showed robust performance when tested on three other publicly available datasets. These results demonstrate the feasibility of robust AI-enabled glaucoma screening.

An auxiliary diagnostic tool for common fundus diseases based on fundus color photography and light-weight classification models.

OBJECTIVE: To evaluate the performance of two lightweight neural network models in the diagnosis of common fundus diseases and make comparison to another two classical models. METHODS: A total of 16,000 color fundus photography were collected, including 2000 each of glaucoma, diabetic retinopathy (DR), high myopia, central retinal vein occlusion (CRVO), age-related macular degeneration (AMD), optic neuropathy, and central serous chorioretinopathy (CSC), in addition to 2000 normal fundus. Fundus photography was obtained from patients or physical examiners who visited the Ophthalmology Department of Beijing Tongren Hospital, Capital Medical University. Each fundus photography has been diagnosed and labeled by two professional ophthalmologists. Two classical classification models (ResNet152 and DenseNet121), and two lightweight classification models (MobileNetV3 and ShufflenetV2), were trained. Area under the curve (AUC), sensitivity, specificity, accuracy, positive predictive value, and negative predictive value were used to evaluate the performance of the four models. RESULTS: Compared with the classical classification model, the total size and number of parameters of the two lightweight classification models were significantly reduced, and the classification speed was sharply improved. Compared with the DenseNet121 model, the ShufflenetV2 model took 50.7% less time to make a diagnosis on a fundus photography. The classical models performed better than lightweight classification models, and Densenet121 showed highest AUC in five out of the seven common fundus diseases. However, the performance of lightweight classification models is satisfying. The AUCs using MobileNetV3 model to diagnose AMD, diabetic retinopathy, glaucoma, CRVO, high myopia, optic atrophy, and CSC were 0.805, 0.892, 0.866, 0.812, 0.887, 0.868, and 0.803, respectively. For ShufflenetV2model, the AUCs for the above seven diseases were 0.856, 0.893, 0.855, 0.884, 0.891, 0.867, and 0.844, respectively. CONCLUSION: The training of light-weight neural network models based on color fundus photography for the diagnosis of common fundus diseases is not only fast but also has a significant reduction in storage size and parameter number compared with the classical classification model, and can achieve satisfactory accuracy.

Deep learning prediction of steep and flat corneal curvature using fundus photography in post-COVID telemedicine era.

Recently, fundus photography (FP) is being increasingly used. Corneal curvature is an essential factor in refractive errors and is associated with several pathological corneal conditions. As FP-based examination systems have already been widely distributed, it would be helpful for telemedicine to extract information such as corneal curvature using FP. This study aims to develop a deep learning model based on FP for corneal curvature prediction by categorizing corneas into steep, regular, and flat groups. The EfficientNetB0 architecture with transfer learning was used to learn FP patterns to predict flat, regular, and steep corneas. In validation, the model achieved a multiclass accuracy of 0.727, a Matthews correlation coefficient of 0.519, and an unweighted Cohen's κ of 0.590. The areas under the receiver operating characteristic curves for binary prediction of flat and steep corneas were 0.863 and 0.848, respectively. The optic nerve and its peripheral areas were the main focus of the model. The developed algorithm shows that FP can potentially be used as an imaging modality to estimate corneal curvature in the post-COVID-19 era, whereby patients may benefit from the detection of abnormal corneal curvatures using FP in the telemedicine setting.