DeepMonitoring: a deep learning-based monitoring system for assessing the quality of cornea images captured by smartphones.

Smartphone-based artificial intelligence (AI) diagnostic systems could assist high-risk patients to self-screen for corneal diseases (e.g., keratitis) instead of detecting them in traditional face-to-face medical practices, enabling the patients to proactively identify their own corneal diseases at an early stage. However, AI diagnostic systems have significantly diminished performance in low-quality images which are unavoidable in real-world environments (especially common in patient-recorded images) due to various factors, hindering the implementation of these systems in clinical practice. Here, we construct a deep learning-based image quality monitoring system (DeepMonitoring) not only to discern low-quality cornea images created by smartphones but also to identify the underlying factors contributing to the generation of such low-quality images, which can guide operators to acquire high-quality images in a timely manner. This system performs well across validation, internal, and external testing sets, with AUCs ranging from 0.984 to 0.999. DeepMonitoring holds the potential to filter out low-quality cornea images produced by smartphones, facilitating the application of smartphone-based AI diagnostic systems in real-world clinical settings, especially in the context of self-screening for corneal diseases.

Innate immune responsive inflammation in development of progressive myopia.

PURPOSE: Inflammation has been implicated for development of myopia. It is not clear when inflammation is kicked in during the course of myopia, and what characteristics of the inflammation. In this study, we tested for cytokines from aqueous humour of eyes with wide spectrum of refractive status for profiling the inflammation. METHODS: Aqueous humour of 142 patient eyes were tested for soluble intercellular adhesion molecule 1 (sICAM-1), monocyte chemoattractant protein-1 (MCP-1), and transforming growth factor-beta 2 (TGF-ß2) using an enzyme-linked immunosorbent assay (ELISA). Eye globe axial length of these patients ranged from emmetropia to high myopia. RESULTS: Of 142 patients, an average axial length is 25.51 ± 3.31 mm, with a range of 21.56-34.37 mm. There are 36 cases in lower 25 percentile, 37 cases in upper 25 percentile, and 69 case in the middle 50 percentile. sICAM-1 and MCP-1 were significantly higher in the eyes with staphyloma (407.48 pg/mL, 312.31 pg/mL, n = 33) or macular schisis (445.86 pg/mL,345.33 pg/mL, n = 19) than that in the eyes without these changes (206.44 pg/mL, 244.76 pg/mL, n = 107). All three cytokines level was significantly associated with eye globe axial in a positive mode while adjusting for the age and sex. Strength of the association was the greatest for sICAM-1 and the weakest for TGF- ß2. MCP-1 was in between. CONCLUSION: sICAM-1 and MCP-1 in ocular fluid may be indicative biomarkers for progressive high myopia and the underneath autoimmune inflammation. sICAM-1 may be used as a monitoring biomarker for development of pathologic myopia.

Artificial intelligence in ophthalmology: The path to the real-world clinic.

Artificial intelligence (AI) has great potential to transform healthcare by enhancing the workflow and productivity of clinicians, enabling existing staff to serve more patients, improving patient outcomes, and reducing health disparities. In the field of ophthalmology, AI systems have shown performance comparable with or even better than experienced ophthalmologists in tasks such as diabetic retinopathy detection and grading. However, despite these quite good results, very few AI systems have been deployed in real-world clinical settings, challenging the true value of these systems. This review provides an overview of the current main AI applications in ophthalmology, describes the challenges that need to be overcome prior to clinical implementation of the AI systems, and discusses the strategies that may pave the way to the clinical translation of these systems.

Effect of Nintedanib Nanothermoreversible Hydrogel on Neovascularization in an Ocular Alkali Burn Rat Model.

PURPOSE: To compare the therapeutic effects of different forms of nintedanib ophthalmic preparations on neovascularization corneal alkali burns in rats. METHODS: Forty rat models of left eye corneal alkali burns were constructed, and the five groups (N = 8) were treated with normal saline, dexamethasone ointment (dexamethasone), 0.2% nintedanib aqueous solution and nintedanib nano thermoreversible hydrogel (NNTH). A slit lamp microscope was used to observe the area of neovascularization. The levels of the inflammatory factors were detected by ELISA. HE staining was performed on the rat corneas. Vascular endothelial growth factor (VEGFA) was detected by immunohistochemistry, and the expression of corneal VEGFA and CD31 was detected by western blotting. An MTT assay was performed to detect the cytotoxicity of nintedanib on human corneal epithelial cells (HCECs) and human umbilical vein vascular endothelial cells (HUVECs). Cell migration was detected by a cell scratch assay, and the proportion of apoptotic cells was detected by Annexin/PI double staining. Immunofluorescence and western blotting were performed to detect the protein expression of VEGFA and CD31. RESULTS: NNTH had a stronger inhibitory effect on corneal neovascularization (CNV) in alkali-burned rats while reducing the level of inflammatory factors. NNTH had a longer drug duration of release than nanoformulations in vitro. Nintedanib at low concentrations (<8 µM) had no significant cytotoxicity to HCECs but significantly induced apoptosis and inhibited the expression of VEGFA and CD31 and the migration of HUVECs. CONCLUSIONS: Nanomorphic thermoreversible hydrogel is superior among the nintedanib ophthalmic preparations, showing better inhibition of CNV in alkali-burned eyeballs and it blocked the migration and proangiogenic ability of HUVECs.

Single-string, closed-loop fixation modification to reposit a dislocated triple-looped haptic intraocular lens.

A technique using the single-string, closed-loop fixation method to reposit dislocated triple-looped haptic intraocular lens (IOL)-capsular bag complex is described. The long needle or curved needle with a 10-0/8-0 polypropylene suture and a 27/30-gauge needle were used as the guide needle to pass through the fenestrated haptics twice. The scleral interlaminar course was used as the fixed point. Last, a fixation knot was created in the sclerotomy by the 2 ends of the thread to close the suture loop for IOL fixation. Another knot was created about 2 to 3 mm from the exit point and was intrasclerally anchored by the aid of the attached needle. 4 eyes from 4 consecutive patients were studied retrospectively; during all follow-up visits, the IOLs were well centered and stable, and no suture erosion, hypotony, scleral atrophy, chronic inflammation, retinal tears, and/or detachments were observed.

A Classification Tree Model with Optical Coherence Tomography Angiography Variables to Screen Early-Stage Diabetic Retinopathy in Diabetic Patients.

AIM: To establish a classification tree model in DR screening and to compare the DR screening accuracy between the classification tree model and the logistic regression model in type 2 diabetes mellitus (T2DM) patients based on OCTA variables. METHODS: Two hundred forty-one eyes of 241 T2DM patients were included and divided into two groups: the development cohort and the validation cohort. Optical coherence tomography angiography (OCTA) images were acquired in these patients. The data of foveal avascular zone area, superficial capillary plexus (SCP) density, and deep capillary plexus (DCP) density were exported after automatically analyzing the macular 6 × 6 mm OCTA images, while the data of radial peripapillary capillary plexus (RPCP) density was exported after automatically analyzing the optic nerve head 4.5 × 4.5 mm OCTA images. These OCTA variables were adopted to establish and validate the logistic regression model and the classification tree model. The area under the curve (AUC), sensitivity, specificity, and statistical power for receiver operating characteristic curves of two models were calculated. RESULTS: In the logistic regression model, best-corrected visual acuity (BCVA) (LogMAR) and SCP density were entered (BVCA : OR= 60.30, 95% CI= [2.40, 1513.82], p = 0.013; SCP density: OR= 0.86, 95% CI= [0.78, 0.96], p = 0.006). The AUC, sensitivity, and specificity for detecting early-stage DR (mild to moderate NPDR) in the development cohort were 0.75 (95% CI: [0.66, 0.85]), 63%, and 83%, respectively. The AUC, sensitivity, and specificity in the validation cohort were 0.75 (95% CI: [0.66, 0.84]), 79%, and 72%, respectively. In the classification tree model, BVCA (LogMAR), DM duration, SCP density, and DCP density were entered. The AUC, sensitivity, and specificity for detecting early-stage DR were 0.72 (95% CI: [0.60, 0.84]), 66%, and 76%, respectively. The AUC, sensitivity, and specificity in the validation cohort were 0.74 (95% CI: [0.65, 0.83]), 74%, and 72%, respectively. The statistical power of the development and validation cohorts in two models was all more than 99%. CONCLUSIONS: Compared to the logistic regression model, the classification tree model has similar accuracy in predicting early-stage DR. The classification tree model with OCTA variables may be a simple tool for clinical practitioners to identify early-stage DR in T2DM patients. Moreover, SCP density is significantly reduced in mild-to-moderate NPDR eyes and might be a biomarker in early-stage DR detection. Further improvement and validation of the DR diagnostic model are awaiting to be performed.

Comparison of deep learning systems and cornea specialists in detecting corneal diseases from low-quality images.

The performance of deep learning in disease detection from high-quality clinical images is identical to and even greater than that of human doctors. However, in low-quality images, deep learning performs poorly. Whether human doctors also have poor performance in low-quality images is unknown. Here, we compared the performance of deep learning systems with that of cornea specialists in detecting corneal diseases from low-quality slit lamp images. The results showed that the cornea specialists performed better than our previously established deep learning system (PEDLS) trained on only high-quality images. The performance of the system trained on both high- and low-quality images was superior to that of the PEDLS while inferior to that of a senior corneal specialist. This study highlights that cornea specialists perform better in low-quality images than the system trained on high-quality images. Adding low-quality images with sufficient diagnostic certainty to the training set can reduce this performance gap.

Preventing corneal blindness caused by keratitis using artificial intelligence.

Keratitis is the main cause of corneal blindness worldwide. Most vision loss caused by keratitis can be avoidable via early detection and treatment. The diagnosis of keratitis often requires skilled ophthalmologists. However, the world is short of ophthalmologists, especially in resource-limited settings, making the early diagnosis of keratitis challenging. Here, we develop a deep learning system for the automated classification of keratitis, other cornea abnormalities, and normal cornea based on 6,567 slit-lamp images. Our system exhibits remarkable performance in cornea images captured by the different types of digital slit lamp cameras and a smartphone with the super macro mode (all AUCs>0.96). The comparable sensitivity and specificity in keratitis detection are observed between the system and experienced cornea specialists. Our system has the potential to be applied to both digital slit lamp cameras and smartphones to promote the early diagnosis and treatment of keratitis, preventing the corneal blindness caused by keratitis.

Development of a deep learning-based image quality control system to detect and filter out ineligible slit-lamp images: A multicenter study.

BACKGROUND AND OBJECTIVE: Previous studies developed artificial intelligence (AI) diagnostic systems only using eligible slit-lamp images for detecting corneal diseases. However, images of ineligible quality (including poor-field, defocused, and poor-location images), which are inevitable in the real world, can cause diagnostic information loss and thus affect downstream AI-based image analysis. Manual evaluation for the eligibility of slit-lamp images often requires an ophthalmologist, and this procedure can be time-consuming and labor-intensive when applied on a large scale. Here, we aimed to develop a deep learning-based image quality control system (DLIQCS) to automatically detect and filter out ineligible slit-lamp images (poor-field, defocused, and poor-location images). METHODS: We developed and externally evaluated the DLIQCS based on 48,530 slit-lamp images (19,890 individuals) that were derived from 4 independent institutions using different types of digital slit lamp cameras. To find the best deep learning model for the DLIQCS, we used 3 algorithms (AlexNet, DenseNet121, and InceptionV3) to train models. The area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and accuracy were leveraged to assess the performance of each algorithm for the classification of poor-field, defocused, poor-location, and eligible images. RESULTS: In an internal test dataset, the best algorithm DenseNet121 had AUCs of 0.999, 1.000, 1.000, and 1.000 in the detection of poor-field, defocused, poor-location, and eligible images, respectively. In external test datasets, the AUCs of the best algorithm DenseNet121 for identifying poor-field, defocused, poor-location, and eligible images were ranged from 0.997 to 0.997, 0.983 to 0.995, 0.995 to 0.998, and 0.999 to 0.999, respectively. CONCLUSIONS: Our DLIQCS can accurately detect poor-field, defocused, poor-location, and eligible slit-lamp images in an automated fashion. This system may serve as a prescreening tool to filter out ineligible images and enable that only eligible images would be transferred to the subsequent AI diagnostic systems.

Diabetes mellitus as a risk factor for retinal vein occlusion: A meta-analysis.

Retinal vein occlusion (RVO) is the second most common etiology for vision loss. There is contrasting evidence on the association between diabetes mellitus (DM) and the risk of RVO. We performed a meta-analysis of published articles before October 31, 2019, to estimate a pooled odds ratio for the association between DM and RVO, including central and branch RVO by a fixed or random effects model. We identified 37 publications from 38 studies (1 publication was from 2 studies), published between 1985 and 2019. In total, 148,654 cases and 23,768,820 controls were included in this meta-analysis. The results of pooled analysis for all 37 publications (or 38 studies) showed a significant association between DM and the risk of RVO (OR = 1.68, 95% CI: 1.43-1.99). Subgroup analysis indicated that DM was significantly associated with CRVO (OR = 1.98, 95% CI: 1.29-3.03, I = 67.9%), but not significantly associated with BRVO (OR = 1.22, 95% CI: 0.95-1.56, I = 64.1%). In conclusion, the result of present meta-analysis suggested that DM is a risk factor for RVO. More well-designed studies on the relationship between RVO and DM should be undertaken in the future.

Inflammatory cytokines in highly myopic eyes.

Currently, myopic retinopathy is the most common irreversible blinding disease but its pathophysiology is not completely clear. A cross-sectional, observational study was conducted in a single center to analyze aqueous samples from highly myopic eyes (axial length >25 mm, n = 92) and ametropic or mild myopic eyes (n = 88) for inflammatory cytokines. Vascular endothelial growth factor (VEGF), Interleukin 6 (IL-6), and matrix metalloproteinase-2 (MMP-2) were measured using an enzyme-linked immunosorbent assay. IL-6 and MMP-2 were significantly higher in the highly myopic eyes than in the non-high myopic eyes (IL-6: 11.90 vs. 4.38 pg/mL, p < 0.0001; MMP-2: 13.10 vs. 8.82 ng/mL, p = 0.0003) while adjusting for age, gender, and intraocular pressure. There was a significant positive association between levels of IL-6 and MMP-2 in aqueous humor and the axial lengths of the eye globes (IL-6, ß = 0.065, p < 0.0001, n = 134; MMP-2, ß = 0.097, p < 0.0001, n = 131). Conversely, VEGF in aqueous humor was significantly lower in the highly myopic eyes than in the non-high myopic eyes (45.56 vs. 96.90 pg/mL, p < 0.0001, n = 153) while age, gender, and intraocular pressure were adjusted. The results suggest that low-grade intraocular inflammation may play an important role in the development and progression of high myopia and myopic retinopathy.

Lysergic acid diethylamide causes photoreceptor cell damage through inducing inflammatory response and oxidative stress.

Lysergic acid diethylamide (LSD), a classical hallucinogen, was used as a popular and notorious substance of abuse in various parts of the world. Its abuse could result in long-lasting abnormalities in retina and little is known about the exact mechanism. This study was to investigate the effect of LSD on macrophage activation state at non-toxic concentration and its resultant toxicity to photoreceptor cells. Results showed that cytotoxicity was caused by LSD on 661 W cells after co-culturing with RAW264.7 cells. Treatment with LSD-induced RAW264.7 cells to the M1 phenotype, releasing more pro-inflammatory cytokines, and increasing the M1-related gene expression. Moreover, after co-culturing with RAW264.7 cells, significant oxidative stress in 661 W cells treated with LSD was observed, by increasing the level of malondialdehyde (MDA) and reactive oxygen species (ROS), and decreasing the level of glutathione (GSH) and the activity of superoxide dismutase (SOD). Our study demonstrated that LSD caused photoreceptor cell damage by inducing inflammatory response and resultant oxidative stress, providing the scientific rationale for the toxicity of LSD to retina.

Metformin inhibits development of diabetic retinopathy through inducing alternative splicing of VEGF-A.

Previous studies have shown that metformin, an AMP-activated protein kinase activator widely prescribed for type 2 diabetes, is especially beneficial in cases of diabetic retinopathy (DR) with undetermined mechanisms. Here, we used a streptozotocin-induced diabetes model in mice to study the effects of metformin on the development of DR. We found that 10 weeks after STZ treatment, DR was induced in STZ-treated mice, regardless treatment of metformin. However, metformin alleviated the DR, seemingly through attenuating the retina neovascularization. The total vascular endothelial cell growth factor A (VEGF-A) in eyes was not altered by metformin, but the phosphorylation of the VEGF receptor 2 (VEGFR2) was decreased, which inhibited VEGF signaling. Further analysis showed that metformin may induce VEGF-A mRNA splicing to VEGF120 isoform to reduce its activation of the VEGFR2. These findings are critical for generating novel medicine for DR treatment.