Ophthalmology and Therapy最新文献

Introduction: The aim of this study was to develop an interpretation pipeline for report generation for slit-lamp (SL) images, using vision-language models (VLMs).

Methods: An image-text alignment module (Bootstrapping Language-Image Pretraining [BLIP]) for report generation was developed using a dataset of SL images paired with medical reports from Zhejiang Provincial People's Hospital (Zhaohui Hospital). These data were split into training and internal validation sets to finetune the BLIP frameworks (LLaVA and Qwen2.5-VL). A dataset from Bijie Hospital was used as an external validation dataset for the evaluation of the frameworks.

Results: The Zhaohui Hospital dataset included 1612 SL images and medical reports. The Bijie Hospital dataset included 100 SL images and medical reports. For the refined LLaVA framework, the BLEU scores (1-4) were 0.560-0.715, ROUGE-L score was 0.731, CIDEr score was 1.712, and SPICE score was 0.329. For the refined Qwen2.5-VL framework, the BLEU scores (1-4) were 0.536-0.696, ROUGE-L score was 0.696, CIDEr score was 1.729, and SPICE score was 0.208. Both models also showed good results in the classification of diseases. The overall accuracy was 0.87 in the refined LLaVA framework and 0.88 in the refined Qwen2.5-VL framework, with high accuracies (≥ 0.9) observed for eyelid disease, pterygium, glaucoma, corneal disease, and conjunctivitis. Interobserver agreement among ophthalmologists was substantial, with κ scores between 0.714 and 0.777. In the evaluation, the 100 reports generated by the LLaVA model demonstrated strong performance across all four metrics-correctness (2.72), completeness (2.79), harmlessness (2.88), and satisfaction (2.73)-with each scoring above 2.7. In comparison, the 100 reports produced by the Qwen model received slightly lower scores than LLaVA in correctness (2.63), completeness (2.70), and satisfaction (2.71).

Conclusions: This study introduced a framework for SL report generation, which enhanced ophthalmic image interpretation and highlighted the potential of VLMs to assist ophthalmologists and patients.

Introduction: Dry eye disease (DED) is a multifactorial inflammatory disorder characterized by tear-film hyperosmolarity, immune activation, and neurosensory dysfunction, which contribute to sustained ocular surface damage. Severe DED is common in autoimmune diseases, especially Sjögren syndrome (SS) and rheumatoid arthritis (RA), and is often refractory to first-line treatments.

Methods: Current evidence on anti-inflammatory therapies was summarized by experts, and the management of challenging cases of autoimmune-related DED followed in different tertiary centers was presented.

Results: Short courses of topical corticosteroids rapidly suppress disease flares and improve clinical signs, including breakup time and ocular surface staining. However, careful stewardship is required, as prolonged use may elevate intraocular pressure, induce cataract formation, and increase infectious risk. For long-term control, immunomodulators such as cyclosporine A (CsA), lifitegrast, and tacrolimus attenuate T-cell-mediated inflammation, promote goblet cell recovery, and stabilize the tear film. Newer CsA formulations have further improved bioavailability and tolerability. Five challenging cases including DED associated with SS or RA, refractory keratopathy, and corneal epithelial defect were described. Management included biological tears, lid-based care, and punctal plugs combined with once-daily CsA, leading to re-epithelialization, symptom relief, and visual stabilization. Adjunctive measures included oral doxycycline to improve meibomian gland function and reduce inflammation. Regular follow-up optimized treatment tapering, safety monitoring, and patient adherence. In two cases, urgent surgical intervention (conjunctival flap, amniotic membrane transplantation, and penetrating keratoplasty) was required.

Conclusions: Autoimmune-related DED requires a stepwise treatment regimen for the stabilization of the ocular surface and the prevention of irreversible damage. This approach involves an initial short course of corticosteroids, followed by sustained immunomodulation (with CsA as the cornerstone), and supplemented by adjunctive therapies targeting meibomian glands and ocular surface epithelium. Multidisciplinary coordination and regular monitoring are essential for maintaining long-term ocular surface homeostasis and satisfactory quality of life and visual function.

Our commentary provides a conceptual foundation for the relationship between ocular half-life and treatment durability with respect to intravitreally administered anti-angiogenic therapies typically used for the treatment of retinal diseases, such as neovascular age-related macular degeneration and diabetic macular edema. Of note, we critically examine claims suggesting that increasing the dose of aflibercept can reduce ocular clearance and, consequently, increase ocular half-life. By reviewing the available data and exploring the underlying pharmacokinetic principles, this commentary seeks to provide an evidence-based understanding of the factors influencing treatment durability in this therapeutic area.

Introduction: This phase 3, multicenter, sham procedure/placebo-controlled, double-masked, randomized clinical trial evaluated the safety and efficacy of oxygen-enriched epithelium-on corneal collagen cross-linking (CXL) for the treatment of keratoconus.

Methods: The trial enrolled patients aged 13 to 51 years diagnosed with keratoconus. Eyes were randomized in a 2:1 ratio to CXL treatment or sham/placebo. The CXL treatment group (200 eyes) received cross-linking using riboflavin 5'-phosphate ophthalmic solutions 0.239% and 0.177% with ultraviolet A (UV-A) irradiation and supplemental oxygen while the sham/placebo group (112 eyes) received placebo solutions and a sham irradiation procedure. The primary efficacy endpoint was the between-group difference in least squares (LS) mean change from baseline in maximum corneal curvature (K_max) at month 12. Safety outcomes included adverse events and ophthalmic assessments.

Results: At month 12, the LS mean K_max in the CXL treatment group improved as a change from baseline by 0.5 D (95% confidence interval [CI] 0.7, 0.3; p < 0.0001) and the LS mean K_max in the untreated sham/placebo group deteriorated by 0.4 D (95% CI 0.1, 0.8: p = 0.0045). The difference between groups was  - 1.0 D (95% CI  - 1.3,  - 0.6; p < 0.0001). Therefore, the study met the primary efficacy endpoint criterion with both a statistically significant and clinically meaningful change in the mean K_max. There were no serious ocular adverse events nor severe treatment-related adverse events in the study eye. The most common adverse event was punctate keratitis (6.5% vs. 1.8% in the CXL and sham/placebo groups, respectively).

Conclusion: Epithelium-on CXL using riboflavin ophthalmic solutions (riboflavin 5'-phosphate ophthalmic solutions 0.239% and 0.177%) with UV-A irradiation and supplemental oxygen is safe and effective for the treatment of keratoconus in pediatric patients and adults.

Trial registration: ClinicalTrials.gov identifier NCT05759559.

Introduction: Patient-reported pain during intravitreal injection (IVI) may vary by injection quadrant and influence adherence to repeated treatments. Hence, this study aimed to quantify whether patient-reported pain differs by IVI quadrant and identify the least-painful approach.

Methods: We searched 13 literature databases on 15 October 2025 for clinical studies reporting patient-reported IVI pain stratified by injection quadrant (superotemporal [ST], superonasal [SN], inferotemporal [IT], inferonasal [IN]). Random-effects meta-analyses synthesized mean differences (MD) in 0-10 visual analogue scale (VAS) pain across all six pairwise quadrant contrasts. All eligible studies were included for a qualitative review and a meta-analysis. The review was prospectively registered (PROSPERO CRD420251164189).

Results: Seven studies comprising 2048 IVI were included (ST 611, SN 469, IT 556, IN 412). Most injections involved anti-VEGF agents. All injections were performed under topical anesthesia (100%; drops or gel), with no subconjunctival anesthesia reported (0%), and pain was recorded immediately post-procedure using VAS-derived scales. Across all six pairwise contrasts, pooled mean differences between quadrants were small (typically < 1 VAS unit on a 0-10 scale), and all 95% confidence intervals crossed zero. Across quadrant comparisons, pooled pain differences were small and inconsistent; none of the pairwise pooled estimates demonstrated a statistically consistent difference, and between-study heterogeneity was substantial to considerable (I² generally > 80%).

Conclusions: Current evidence does not demonstrate a clearly less painful quadrant for intravitreal injections. Estimates varied across studies and heterogeneity was substantial, limiting confidence in any single average quadrant effect. Efforts to minimize IVI-related pain should prioritize optimized anesthesia protocols, communication, and injection technique, while future studies should standardize pain measurement and quadrant allocation to clarify whether subtle quadrant effects exist.

Trial registration: PROSPERO registration: CRD420251164189.

Introduction: To validate the diagnostic performance of the Eyerobo FC, a new portable non-mydriatic fundus camera for diabetic retinopathy (DR) screening, against an established desktop fundus camera benchmark using a transfer-learning approach in which artificial intelligence (AI)-based detection algorithms trained on desktop images were applied to Eyerobo FC images.

Methods: This prospective validation study employed a three-tier experimental design. Tier 1 involved training a deep learning model (EfficientNet-B4) on standard desktop camera images from EyePACS and APTOS 2019 datasets. Tier 2 established the reference standard by evaluating the trained model on the Messidor-2 dataset (N = 1748 eyes) captured with a Topcon TRC NW6 desktop camera (sensitivity 92.7%, 95% CI 91.2-94.2%; AUC 0.952, 95% CI 0.943-0.961). Tier 3 validated the same AI model (without retraining) on images from the Eyerobo FC in a prospective cohort (N = 104 eyes: 52 referable DR, 52 non-referable). The primary outcome was noninferiority of sensitivity and specificity (margin $\Delta$ = 10%) compared with the desktop benchmark. Statistical analysis included bootstrap resampling (1000 iterations) for confidence intervals and a one-sided Z-test for the difference of proportions to assess noninferiority.

Results: In a balanced cohort of 104 eyes (52 referable DR, 52 non-referable), the Eyerobo FC achieved sensitivity of 92.3% (95% CI 84.4-98.2%) and specificity of 94.2% (95% CI 87.0-100%), demonstrating noninferior performance compared with the desktop benchmark (sensitivity 92.7%, specificity 94.3%). The sensitivity difference of $-0.4$ percentage points and the specificity difference of $-0.1$ percentage points were both within the noninferiority margin. AUC was 0.977 (95% CI 0.945-0.997) versus 0.952 for the desktop benchmark. The AI model correctly classified 97 of 104 eyes (93.3% accuracy, 95% CI 88.5-98.1%), with 4 false negatives and 3 false positives. Noninferiority was statistically confirmed for both sensitivity and specificity (P < 0.05). Inter-grader agreement was excellent (Cohen's kappa = 0.917). Nonmydriatic image gradability rate was 94.4%. Grad-CAM visualization confirmed appropriate model attention to hemorrhages, exudates, and microaneurysms rather than artifacts.

Conclusions: The Eyerobo fundus camera demonstrates noninferior diagnostic performance (sensitivity 92.3%, specificity 94.2%, AUC 0.977) compared with desktop systems when evaluated with AI algorithms trained exclusively on desktop images. These findings support deploying portable AI-assisted screening in resource-constrained and point-of-care settings, with successful cross-domain transfer learning enabling algorithmic generalizability across imaging platforms.

Introduction: This study aimed to assess the tomographic and biomechanical progression of non-operated eyes in patients with asymmetric keratoconus who underwent unilateral intracorneal ring segment (ICRS) implantation.

Methods: This retrospective study included 33 patients (66 eyes) with asymmetric keratoconus who underwent unilateral ICRS implantation. The fellow eye did not meet clinical or tomographic criteria for surgical intervention at the time of the initial surgery. Tomographic and biomechanical data from the non-operated fellow eye were compared between baseline and the final follow-up (minimum 12 months). Scheimpflug-based corneal tomography (Pentacam^®) assessed keratometry, curvature, corneal irregularity indices, and thickness-based metrics. Biomechanics parameters were evaluated via dynamic corneal response analysis (Corvis ST^®). Disease progression was defined as the presence of at least two concordant changes among the following: Kmax increase > 1.0 D, thinnest corneal thickness decrease > 20 µm, BAD-D increase > 0.42, and worsening of biomechanical E-stage.

Results: Over a median follow-up of 31 months, non-operated fellow eyes showed no clinically meaningful changes in visual, tomographic, or biomechanical parameters, suggesting no structural progression under conservative, behavior-based management, including counseling to avoid eye rubbing, management of atopic disease, and oral riboflavin supplementation. Disease progression was observed in only two eyes (6.1%), which were subsequently referred for corneal cross-linking (CXL). ICRS-treated eyes demonstrated significant improvements in visual acuity and corneal regularity.

Conclusion: In this paired-eye study, most non-operated fellow eyes in asymmetric keratoconus remained stable under conservative, behavior-based management combined with regular multimodal monitoring. These results support a behavior-first approach, using corneal tomography and biomechanics to detect objective progression and guide timely intervention. Conservative management in asymmetric keratoconus should be understood as an active, follow-up-dependent strategy rather than the absence of treatment. Surgical interventions should remain reserved for objectively progressing cases, recognizing that a small but clinically relevant proportion of fellow eyes still progress and require CXL.

Receiving a diagnosis of a vision-threatening eye condition is a psychologically salient event for many patients. Across ophthalmology, fear, uncertainty, and emotional distress are commonly observed at the time of diagnosis, often before substantial functional vision loss occurs. Although depression and anxiety are well documented in chronic eye disease, diagnosis-linked distress remains inconsistently recognised and is rarely addressed systematically within routine ophthalmic care. This article introduces post-visual diagnosis distress (PVDD) as a descriptive, nondiagnostic framework to characterise the predictable, loss-related emotional responses that may arise following the diagnosis of a vision-threatening or irreversible eye condition. PVDD is not proposed as a psychiatric disorder, but as a clinically meaningful construct that helps legitimise patient experience, supports empathetic communication, and facilitates timely recognition and appropriate signposting when distress is significant. Drawing on evidence from age-related macular degeneration, glaucoma, and other chronic eye diseases, and informed by multidisciplinary perspectives spanning ophthalmology, psychiatry, counselling, patient advocacy, and public health, the article argues that recognising PVDD may improve patient engagement, quality of life, and psychological outcomes.

Introduction: The aim of this study was to assess the effect of including lens thickness (LT) in modern intraocular lens (IOL) formulas and evaluate the influence of ocular biometric parameters on IOL power calculation accuracy in short and normal axial length (AL) eyes.

Methods: This retrospective cohort study included 174 eyes (74 short AL < 22 mm, 100 normal AL 22-24 mm). Preoperative biometry was obtained using IOLMaster 700. Predicted refractions were calculated using SRK/T, Haigis, Barrett Universal II (BUII), Kane, and Emmetropia Verifying Optical (EVO) formulas, with and without LT. Spearman correlation and multivariate regression analyses were performed to identify biometric predictors of refractive error.

Results: Short eyes had significantly shallower anterior chamber depth (ACD), smaller white-to-white (WTW), thicker LT, and higher mean keratometry (Km) and corneal astigmatism (CA) (all p < 0.01). No significant differences in mean absolute error (MAE) were detected between most formula pairs after adjustment in either subgroup. LT inclusion resulted in a less myopic mean prediction error in both subgroups (all p < 0.01), but MAE improvements were only seen in normal eyes (p < 0.05). AL and WTW were consistent predictors of absolute prediction error (APE) overall, with shorter AL and smaller WTW associated with greater APE. In short eyes, AL showed the strongest association with APE, whereas in normal eyes, WTW was most strongly correlated with APE.

Conclusions: Short eyes exhibited greater biometric variability. Incorporation of LT improved the overall prediction accuracy of modern IOL formulas, particularly in normal AL eyes. AL and WTW were the most influential biometric factors affecting refractive prediction outcomes.

Trial registration: Chinese Clinical Trial Registry (ChiCTR), ChiCTR2600116749; retrospectively registered on 14 January 2026.

Introduction: To examine the current landscape of artificial intelligence (AI) applications in vitreoretinal (VR) diseases and surgery, with the aim of identifying knowledge gaps and guiding future directions in this rapidly evolving field.

Methods: Systematic review including original studies involving the use of AI and focusing on VR pathologies. A comprehensive electronic search of the literature was carried out in multiple databases.

Results: Thirty-seven studies were included. Most evaluated machine learning or deep learning models for preoperative prognostication using optical coherence tomography with or without clinical variables. Predictive performance for postoperative best-corrected visual acuity (BCVA) was high in several cohorts (R² up to 0.80; area under the receiver operating characteristic curve [AUROC] > 0.95), with models consistently highlighting outer retinal biomarkers as key determinants of visual recovery after epiretinal membrane and macular hole surgery. For anatomical outcomes, deep learning models frequently achieved > 90% accuracy in predicting macular hole closure and retinal reattachment/reattachment-related endpoints. Intraoperative computer-vision systems demonstrated feasibility for real-time instrument detection and tracking, reporting precision above 90% in experimental or early clinical settings. Large language models showed moderate-to-high agreement with expert surgical planning (80-93%) and potential utility in education and workflow support; however, across domains, most studies were retrospective and single-center, with limited external validation.

Conclusions: AI may transform vitreoretinal surgery, from outcome prediction to intraoperative guidance and workflow support. Despite strong performance in research settings, broader clinical adoption requires prospective validation to ensure reliability, transparency, and real-world benefit.