To investigate how different light/dark (L/D) cycles and light intensities influence refractive development and form-deprivation myopia (FDM) in mice, and to examine whether these phenotypes are accompanied by changes in retina/RPE circadian rhythm-related genes expression and dopamine levels. Three-week-old C57BL/6 J mice undergoing natural refractive development or monocular FDM (induced by translucent diffusers) were exposed for 4 weeks to three light intensities (10, 300, or 3000 lx) under four light/dark cycles. Refractive error, axial length (AL), Opn4, Cry1, Per3 expression, and dopamine levels in retina/RPE were assessed. Extending photoperiod (from 6L/18D to 12L/12D, 18L/6D, or continuous light) induced significant myopic shifts (increased AL, decreased refraction) in natural refractive development mice at 300 lx and 3000 lx (both P < 0.001), but not at 10 lx. This shift correlated with decreased Opn4 and increased Cry1 and Per3 expression (P < 0.05). Notably, continuous light (CL) at 3000 lx significantly inhibited FDM progression (vs. other photoperiods P < 0.001), accompanied by elevated Opn4 and Cry1. Conversely, one week of CL at 10 lx transiently induced hyperopia in FDM mice. Retinal dopamine content showed no significant correlation with these refractive changes. Light intensity and L/D cycles critically regulate refractive development and myopia progression in mice, with these effects showing stronger associations with alterations in circadian rhythm-related genes than with retinal dopamine levels. Continuous bright light (3000 lx) inhibits FDM, while dim light (10 lx) effects are photoperiod dependent. These findings suggest targeting optimized light exposure and circadian modulation as potential therapeutic strategies.

This multicentre longitudinal study investigated nasotemporal retinal asymmetry as a geometric biomarker for monitoring myopia risk in children. A total of 2520 Chinese children (Beijing cohort: N = 49, 5–14 years old; Anyang cohort: N = 2471, 6–9 years old) underwent annual measurements over 2 years, including axial length (AL), central spherical equivalent (CSE) and peripheral AL (30° nasal, temporal, superior, inferior) in the Beijing cohort and horizontal meridian peripheral refractions (±15°, ±30°) in the Anyang cohort. Retinal morphology was quantified through the vertex radius of curvature, asphericity and areas under the horizontal nasal and temporal retinal curve (AUHRCn and AUHRCt, respectively). Nasotemporal retinal asymmetry was defined as AUHRCt/n. Vertical asymmetry was defined as superior peripheral AL/inferior peripheral AL. Participants from the Anyang cohort were stratified into persistent myopia (myopic at baseline), newly developed myopia (myopia onset during follow-up) and persistent non-myopia (remained non-myopic over 2 years). In the Beijing cohort, the temporal-to-nasal AL ratio was the only factor significantly associated with myopic progression (ΔAL: r = −0.36, p = 0.01; ΔCSE: r = 0.37, p = 0.01). Validating this, in the Anyang cohort, decreasing AUHRCn preceded myopia onset, followed by progressive temporal steepening. Linear regression identified baseline AUHRCt/n as an independent negative predictor of myopic progression (β = 2.90, p = 0.03). A model incorporating baseline AUHRCt/n improved the prediction of progression risk (integrated discrimination improvement (IDI) = 3.74

PURPOSE:This study investigated the heterogeneity and molecular alterations of trabecular meshwork (TM) in glaucomatous non-human primates, aiming to identify potential pathogenic genes and signaling pathways involved in glaucoma. METHODS:TM tissues were analyzed using 10x Genomics single-cell RNA sequencing on the Illumina platform. After quality control, dimensionality reduction and clustering were performed. Differentially expressed genes (DEGs) were identified using Seurat and DESeq2. Functional enrichment analyses were conducted via Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG). Protein expression was assessed by Western blotting, mRNA levels by RT-PCR, protein localization and fluorescence intensity by immunofluorescence, and mitochondrial morphology by transmission electron microscopy. RESULTS:We classified the outflow tissue into 10 cell subtypes, and TM cells were further classified into three subtypes: Beam A, Beam B, and JCT. Beam A cells exhibited the most significant proportional changes, and GO analysis indicated their strong association with extracellular matrix regulation. MGP emerged as the most significantly upregulated gene in both the overall TM population and the Beam A subcluster. Dexamethasone (DEX) treatment induced MGP expression in human primary TM cells, while MGP knockdown partially restored Wnt/β-catenin signaling, decreased Wnt5a levels, and reduced ECM deposition. CONCLUSION:This study generated a transcriptional atlas of TM in glaucoma. MGP may contribute to TM dysfunction via the Wnt/β-catenin-Wnt5a-ECM signaling pathway, offering novel insights into potential therapeutic targets for glaucoma intervention.

Wearable silent speech systems hold significant potential for restoring communication in patients with speech impairments. However, seamless, coherent speech remains elusive, and clinical efficacy is still unproven. Here, we present an AI-driven intelligent throat (IT) system that integrates throat muscle vibrations and carotid pulse signal sensors with large language model (LLM) processing to enable fluent, emotionally expressive communication. The system utilizes ultrasensitive textile strain sensors to capture high-quality signals from the neck area and supports token-level processing for real-time, continuous speech decoding, enabling seamless, delay-free communication. In tests with five stroke patients with dysarthria, IT's LLM agents intelligently corrected token errors and enriched sentence-level emotional and logical coherence, achieving low error rates (4.2% word error rate, 2.9% sentence error rate) and a 55% increase in user satisfaction. This work establishes a portable, intuitive communication platform for patients with dysarthria with the potential to be applied broadly across different neurological conditions and in multi-language support systems.

Aims To investigate the association between visual impairment (VI) and cognitive impairment in middle-aged and older adults from a rural Chinese population. Methods The Handan Eye Follow-up Study was conducted between 2012 and 2013. All participants underwent a standardised ophthalmic examination, structured interview and mobility performance assessment. The ophthalmic examination included visual acuity (VA) measurement and a comprehensive clinical eye examination. The interview collected demographic characteristics, cognitive status assessed by the Mini-Mental State Examination (MMSE) and general information related to medical conditions. Mobility performance was evaluated using walking speed tests, standing balance tests and the Timed Up and Go Test. Results Of the 5394 participants in the Handan Eye Follow-up Study, 3794 were aged ≥50 years, of whom 3650 completed both VA and cognitive assessments. The prevalence of distance VI and near VI was 27.53% and 91.99%, respectively. Linear regression analyses revealed that distance VI was associated with a 1.53-fold higher likelihood of poorer MMSE scores after adjustment for demographic variables (p<0.01); however, this association was not statistically significant after further adjustment for all covariates (p=0.23). Near VI was associated with a 2.21-fold higher likelihood of poorer MMSE scores following initial adjustment (p=0.025), and this association remained significant after full adjustment (OR= 3.44, p=0.028). Conclusions Near VI was associated with poorer cognitive performance in a rural Chinese population of middle-aged and older adults, highlighting the potential role of correctable vision loss in cognitive health.

Light is a major environmental signal that shapes circadian rhythms, mood regulation, and ocular growth through a network of non-visual photoreceptive pathways. Increasing evidence suggests that photic information, particularly as decoded by intrinsically photosensitive retinal ganglion cells (ipRGCs), converges on central circuits governing both affective states and refractive development. To integrate these cross-system interactions, we propose the conceptual framework of a “light–eye–brain axis,” which outlines how environmental light cues are encoded by the retina and subsequently modulate neuroendocrine, autonomic, and inflammatory processes. Within this framework, mood disturbances may contribute to myopic progression through altered light-exposure behaviors, neurotransmitter imbalance, hypothalamic–pituitary–adrenal axis instability, and impaired neuroplasticity, whereas high myopia may increase vulnerability to anxiety or depressive symptoms through shared neural and immune pathways. Taken together, this integrative perspective highlights how light-dependent signaling shapes both emotional and refractive outcomes, and provides a conceptual foundation for future mechanistic studies as well as evidence-informed approaches to optimizing light exposure in the context of mood and visual health.

To validate a wrist-worn multimodal wearable system for synchronised monitoring of multispectral light exposure, activity phenotypes and five-stage sleep architecture in children and adolescents in a structured 24-h pilot validation protocol. A wrist-mounted multimodal wearable system was developed, integrating a 9-channel multispectral light sensor (350–1000 nm), a 9-axis inertial measurement unit (IMU) and a photoplethysmography (PPG) sensor for synchronised acquisition of environmental light, body movement and cardiovascular signals. Twelve healthy children and adolescents (mean age: 10.04 ± 3.26 years; range: 6–15 years) underwent a 24-h pilot validation protocol. Long short-term memory (LSTM) networks were developed for activity recognition and a Stacking ensemble framework combined with cost-sensitive learning was implemented for five-stage sleep classification, i.e., wake (W), non-rapid eye movement stage 1 (N1), non-rapid eye movement stage 2 (N2), non-rapid eye movement stage 3 (N3) and rapid eye movement (REM). Under the three lighting conditions tested, the device captured distinct spectral irradiance profiles: outdoor natural light (1.8 W m−2 nm−1; broad-spectrum 350–1000 nm), indoor natural light (0.30 W m−2 nm−1) and artificial illumination (0.07 W m−2 nm−1; peak at 545–600 nm). The LSTM-based model achieved 86.33

AIMS:To investigate retinal vascular geometric alterations following intraocular pressure (IOP)-lowering surgery in patients with primary angle closure disease (PACD) and determine associations between IOP reduction magnitude and retinal vascular parameters. METHODS:This retrospective, self-controlled study included patients with PACD who underwent IOP-lowering surgery at Beijing Tongren Hospital. Retinal vascular parameters were quantitatively measured from fundus photographs before and after surgery using a validated deep learning-based analysis system. Main outcomes included retinal arteriolar and venular calibre, tortuosity, fractal dimension (FD), segment number (N_seg) and vascular density. Linear mixed-effects models, adjusted for age, follow-up interval and PACD subtype, assessed associations between IOP reduction magnitude and vascular parameter changes. RESULTS:Among 126 eyes (112 patients), surgery decreased IOP from 35.0±9.9 to 16.7±5.7 mm Hg (p<0.001). Venous FD increased from 1.52±0.06 to 1.53±0.05 (p=0.003), N_seg from 123.50 (IQR: 61.75-182.50) to 153.00 (88.50-185.00) (p=0.007), vessel area density (VAD) from 6.13±1.56 to 6.50±1.34 (p=0.006) and vessel skeleton density (VSD) from 1.24±0.35 to 1.35±0.31 (p=0.003). Vessel calibre and tortuosity showed no significant changes. IOP reduction magnitude was independently associated with changes in venous FD (p=0.023), N_seg (p=0.031), VAD (p=0.012) and VSD (p=0.030). CONCLUSIONS:Surgical IOP reduction in PACD patients produces selective retinal venous alterations with increased complexity and density. These pressure-responsive changes may provide complementary, non-invasive information regarding vascular response after surgery.

Myopia poses a growing global public health challenge, with its underlying mechanisms not yet fully understood. While insufficient daytime light exposure is a recognized protective factor, the role of nighttime light exposure remains contentious. Through a murine model, we uncover a novel melanopsin-dependent pathway by which light-at-night (LAN) exposure disrupts the expression of core circadian clock genes (including Arntl, Cry1, Per1, and Per2) and downregulates melanopsin expression in intrinsically photosensitive retinal ganglion cells (ipRGCs). These alterations are associated with axial elongation. Crucially, melanopsin-knockout mice exhibit resistance to myopic changes induced by LAN, indicating that melanopsin-dependent pathways mediate ocular growth responses. Supporting these experimental findings, our epidemiological analysis demonstrates that LAN exposure significantly increases the risk of incident myopia onset in humans. Together, these results establish a functional link between environmental light exposure and the molecular pathogenesis of myopia, identifying LAN as a modifiable risk factor for myopia. Our study provides important insights into the light-mediated myopia development and suggests melanopsin signaling as a potential target for preventive and therapeutic strategies.

Disruption of lipid metabolism occurs in both the retina and optic nerve following optic nerve crush (ONC); however, whether the regulation of lipid droplets (LDs) influences the survival and function of retinal ganglion cells (RGCs) remains largely elusive. Here, we identified the distribution of LDs in the retina and optic nerve of ONC mice, and A922500, a selective inhibitor of diacylglycerol acyltransferase 1 (DGAT1), was applied to investigate the effect of suppressing LD biogenesis on RGC survival and visual function. The results showed that LDs were localized in Tuj1-positive cells within the retina and in IBA1-positive cells within GFAP-negative region of optic nerve following injury. A922500 significantly reduced LD accumulation and attenuated microglial density in both retina and optic nerve, but did not affect RGC survival following injury. No regenerating CTB-labeled axons were observed with A922500 monotherapy. However, combined treatment with A922500 and AAV2-CNTF enhanced axon regeneration compared to CNTF-treated alone. A922500 also significantly improved flash visual evoked potential recordings in both ONC control mice and CNTF-treated mice after injury. This study demonstrates that optic nerve crush induces LD accumulation in retinal RGC somas and in microglia at the optic nerve injury site, and reducing LDs does not preserve RGC survival but attenuates injury response, enhances CNTF-induced axon regeneration, and improves visual function.

Abstract Background This study aimed to observe the safety and effectiveness of combined phacoemulsification and goniosynechialysis (PEI-GSL) in patients with primary angle-closure glaucoma (PACG) and coexisting cataracts. Methods This multicenter prospective study enrolled 289 consecutive patients with PACG and cataracts who were treated using PEI-GSL between November 1, 2022 and March 12, 2024. All patients underwent comprehensive ocular evaluations at baseline, at 1 week, and at 1, 3, 6, and 12 months postoperatively. Complete success was defined as achieving an intraocular pressure (IOP) of 6–18 mmHg without antiglaucoma medications or reoperation. Qualified success was similarly defined but allowed antiglaucoma medications. Results Among patients with PACG, 237 (82.0%) of 289 completed a 12-month follow-up. The baseline IOP decreased from 22.9 ± 10.2 to 14.0 ± 3.1 mmHg at 12 months, and the number of antiglaucoma medications declined from 2.08 ± 1.23 to 0.14 ± 0.52. The complete and qualified success rates were 80.6% and 88.3%, respectively. Postoperative complications occurred in 49 (17.0%) patients; while 3 (1.0%) patients underwent yttrium-aluminum-garnet (YAG) laser capsulotomy and 6 (2.1%) required glaucoma surgery. Multivariate Cox regression analysis revealed that the anterior chamber depth ( P = 0.044) and axial length ( P = 0.050) were positively associated with qualified success. The complete and qualified success rates among patients with advanced PACG were 78.2% and 87.9%, respectively. Conclusion We demonstrated that PEI-GSL is an effective and safe treatment for patients with PACG and coexisting cataracts, even in advanced stages.

Purpose:The purpose of this study was to investigate the effects of repeated red laser irradiation at a supraclinical dose on histology and apoptosis-related molecular markers in guinea pig ocular tissues. Methods:Guinea pigs (1 month old) were randomized into two groups: the repeated irradiation group and the control group. After pupil dilation, the repeated irradiation group received binocular red laser (650 nm, 10 mW) exposure for 3 minutes per session, 6 sessions daily, over 8 days. Ocular surface morphology was examined using slit lamp microscopy. Retinal histopathology was assessed by hematoxylin and eosin (H&E) staining, terminal deoxynucleotidyl transferase dUTP nick-end labeling (TUNEL) assay, immunohistochemistry, and immunofluorescence. Protein and mRNA expression of apoptosis-related markers (Caspase-3, Caspase-9, Bax, and Bcl-2) were quantified by Western blotting and quantitative polymerase chain reaction. Results:Repeated red laser exposure-induced lesions localized to the retina temporal to optic nerve, characterized by outer nuclear layer (ONL) pyknosis, photoreceptor inner/outer segments degeneration, and retinal pigment epithelium (RPE) disruption. TUNEL staining revealed a significant increase in apoptotic cells within the ONL of irradiated eyes compared with controls (84.33 ± 54.32 vs. 5.06 ± 2.99 cells/section, P = 0.002). Analysis of apoptosis-related proteins revealed upregulation of Bax protein (P < 0.05), downregulation of Bcl-2 mRNA (P < 0.05), and an increased Bax/Bcl-2 mRNA ratio (P < 0.01). Pro-Caspase-9 protein levels decreased (P < 0.05), whereas Caspase-3 gene expression was elevated (P < 0.001). The RPE layer showed disrupted continuity with RPE65 expression reduced within 1000 µm of the optic nerve (1.00 ± 0.13 vs. 0.69 ± 0.07, P < 0.001). Conclusions:Through repeated red laser irradiation at supraclinical dose and frequencies, this study identified the vulnerable targets of red laser in the fundus, demonstrating damage to the ONL, photoreceptor inner/outer segments, and RPE, whereas no adverse effects were observed on anterior segment structures including the cornea and lens.

BACKGROUND:Cataract is the leading cause of blindness worldwide despite treatment being very cost-effective. To stimulate efforts to address the large unmet need for affordable, good-quality cataract services, member states at the 74th World Health Assembly endorsed a global target of a 30 percentage-point increase in effective cataract surgical coverage (eCSC) by 2030. In this Article, we report on progress towards that target. METHODS:We did a secondary analysis of 233 participant-level population-based survey datasets (Rapid Assessment of Avoidable Blindness [RAAB] surveys and other population-based cross-sectional surveys) from 68 countries (2003-24) to estimate eCSC with a cataract surgical threshold of worse than 6/18 pinhole visual acuity and a threshold of 6/18 or better for postoperative presenting visual acuity (eCSC6/18) for each country. Eligible studies included people aged 50 years and older, without any further age restrictions. eCSC was calculated as the number of people who had received surgery and had a good visual outcome as a proportion of all people who had received or were still to receive surgery. For countries with only one survey estimate available, we reported the survey as the country estimate; for each country with two or more surveys, we used more recent and nationally representative survey to generate estimates. We included 228 of 233 datasets (excluding five datasets from the European region) in a mixed-effects logistic regression to estimate global and WHO regional temporal trends in eCSC6/18 from 2000 to 2030. For global estimates, we used eCSC in the Americas region as a proxy for the European region. We used 120 RAAB surveys that measured visual acuity at the 6/12 threshold to analyse causes of non-good outcomes (postoperative presenting visual acuity worse than 6/12) and estimated potential gains in eCSC (with a cataract surgical threshold of worse than 6/12 pinhole visual acuity and a 6/12 or better threshold for postoperative presenting visual acuity [eCSC6/12]) from correcting residual postoperative refractive errors. Potential gains were estimated by comparing eCSC6/12 calculated using postoperative presenting visual acuity with eCSC6/12 calculated using pinhole visual acuity (a proxy for visual acuity with optical correction of refractive error). FINDINGS:We used 130 studies to report 68 country estimates of eCSC6/18. Country estimates for eCSC6/18 from surveys ranged from 2·1% (95% CI 0·9-3·4) in Burundi (2024) to 77·7% (72·9-82·5) in Qatar (2023). Globally, we predicted eCSC6/18 of 48·2% (39·7-57·2) in 2025, increasing by 8·4 percentage points (8·1-8·6) between 2020 and 2030 (from 43·9% [36·6-51·2] to 52·3% [41·4-62·7]). Uncorrected refractive error accounted for a median 26·4% of non-good outcomes per survey; we estimated that treating this condition could yield a median 3·7 percentage-point gain in eCSC6/12 globally. INTERPRETATION:Global efforts are not on track to meet the target of a 30 percentage-point increase in eCSC between 2020 and 2030-immediate and substantial investment in cataract services is, therefore, required. Survey data indicated which elements of the cataract care pathway could be targeted for quality-improvement initiatives. A suite of cataract surgery quality indicators should be collected to better inform service providers. FUNDING:Indigo Trust, The Fred Hollows Foundation, and Wellcome Trust.

The trabecular meshwork (TM) plays a pivotal role in maintaining intraocular pressure (IOP) by regulating aqueous humor outflow. Nuclear factor erythroid 2-related factor 2 (NRF2) was identified as a key transcriptional controller of TM redox balance and mitochondrial function. Transcriptomic profiling of tert-butyl hydroperoxide (tBHP)-induced oxidative injury revealed NRF2 pathway involvement in TM cellular defense. NRF2 knockout (KO) mice exhibited impaired aqueous humor dynamics, elevated IOP, and TM oxidative damage. In vitro, NRF2 knockdown aggravated oxidative stress and mitochondrial dysfunction, whereas NRF2 overexpression mitigated tBHP-induced cytotoxicity. The results of the gene set enrichment analysis (GSEA) indicated enrichment of oxidative phosphorylation pathway in NRF2-deficient cells. Chromatin immunoprecipitation sequencing (ChIP-seq) confirmed NDUFS7 as a direct NRF2 target essential for mitochondrial complex I integrity. Restoration of NDUFS7 expression in NRF2-deficient TM cells or KO mice rescued mitochondrial impairment. Collectively, these findings establish the NRF2/NDUFS7 axis as a central defense mechanism protecting TM from oxidative injury and suggest potential therapeutic strategies for glaucoma-associated ocular hypertension.

Importance:Childhood myopia is a global health concern with escalating prevalence and can lead to severe irreversible visual impairment. Early prediction of myopia progression is crucial for timely intervention to prevent high myopia and associated complications. Objective:To develop and validate a quantitative method, based on a deep learning method and using only fundus images and baseline refraction data, to predict both myopia progression trajectory and high myopia risk in schoolchildren. Design, Setting, and Participants:This longitudinal school-based cohort study (Anyang Childhood Eye Study) was conducted from February 2012 to May 2018, with annual follow-up examinations from February 2013 to May 2018. Grade 1 students, aged 6 to 9 years, were recruited from 11 randomly selected urban primary schools in Anyang, Henan Province, China. Children who received myopia control treatments, had amblyopia, or underwent strabismus surgery were excluded. Two independent external validation cohorts were used: the Lhasa cohort (predominantly consisting of Tibetan children) and the Beijing cohort (predominantly consisting of Han Chinese children). Data analysis was performed from July 2024 to July 2025. Main Outcomes and Measures:Performance of a novel deep learning model, created from combining convolutional neural network (34-layer residual network) and recurrent neural network (long short-term memory network). Area under the curve (AUC) was used to assess myopia and high myopia risk prediction, and mean absolute error (MAE) was used to assess spherical equivalent refraction (SER) prediction. Myopia was defined as SER of -0.5 D or less, and high myopia was defined as -6.0 D or less, using cycloplegic autorefraction. Results:Among the 3048 children (mean [SD] age, 7.1 [0.4] years; 1716 females [56.3%]) included, the baseline prevalence rate of myopia was 5.71% (174) and high myopia was 0.5% (15). The deep learning model achieved AUC scores of 0.941 (95% CI, 0.936-0.946) for myopia risk prediction and 0.985 (95% CI, 0.982-0.988) for high myopia risk prediction, with an overall MAE of 0.322 D per year for SER prediction. External validation in the Beijing cohort (n = 130; mean [SD] age, 9.9 [3.7] years; 82 males [63.1%]; 128 Han Chinese [98.5%], 2 Man Chu [1.5%]) and the Lhasa cohort (n = 1039; mean [SD] age, 6.8 [0.5] years; 536 males [51.6%]; 29 Han Chinese [2.8%], 1007 Tibetan [96.9%], 3 [0.2%] from other ethnic minority groups) demonstrated maintained cross-ethnicity performance by the model, with an MAE of 0.355 D and 0.261 D per year, respectively. Conclusions and Relevance:In this cohort study, a deep learning model, using minimal baseline data, provided highly accurate prediction of myopia and high myopia risk. This deep learning approach suggests the utility of large-scale screening and early-intervention efforts in resource-limited settings.

Purpose:To explore a stepwise atropine treatment strategy for myopic control: beginning with 0.01% atropine for 6 months, followed by adjustment to higher concentrations based on treatment response. Design:Prospective study. Participants:Total of 150 children aged 8 to 14 years with spherical equivalent refractive error (SER) range of -1.00 D to -6.00 D in both eyes were enrolled. Methods:The study was divided into 2 phases. In the initial 6-month period, children received 0.01% atropine eye drops daily in both eyes. In the subsequent 12-month switchover period, children with less than -0.25 D progression continued with 0.01% atropine; those with progression between -0.25 D and -0.375 D were shifted to 0.02% atropine; and those with progression over -0.375 D were switched to 0.04% atropine. Cycloplegic refraction, axial length (AL), pupil size, and intraocular pressure were measured at baseline, 6 months, 12 months, and 18 months. Main Outcome Measures:Differences in myopia progression and axial elongation from 6 months to 18 months among 3 treatment groups. Results:In the initial 6-month period, mean SER changes for the 0.01% atropine groups, 0.02%, and 0.04% atropine switchover groups were -0.03 ± 0.18 D, -0.31 ± 0.07 D, and -0.53 ± 0.12 D, respectively, which reflect the group assignment rule. In the 0.02% atropine switchover group, myopia progression reduced from -0.31 ± 0.07 D to -0.09 ± 0.22 D (6-12 months) and -0.12 ± 0.25 D (12-18 months). The 0.04% atropine switchover group showed a reduction from -0.53 ± 0.12 D to -0.02 ± 0.18 D (6-12 months) and -0.16 ± 0.31 D (12-18 months). Between 6 months and 18 months of subsequent 12-month switchover period, the mean SER change was -0.31 ± 0.30 D, -0.21 ± 0.38 D, and -0.18 ± 0.40 D in the 0.01%, 0.02%, and 0.04% atropine groups, respectively, with a respective mean increase in AL of 0.15 ± 0.11 mm, 0.13 ± 0.11 mm, and 0.14 ± 0.10 mm. Conclusions:This study provides preliminary evidence that a stepwise atropine treatment regimen, starting with 0.01% atropine for 6 months and then increasing to 0.02% or 0.04% based on myopia progression, appears to be a promising strategy that warrants further investigation in controlled trials. Financial Disclosures:Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

Understanding causal mechanisms in deep learning is essential for clinical adoption, where interpretability and reliability are critical. Most existing AI systems act as black boxes, limiting transparency in medicine. We propose a causal inference framework integrated into neural networks to assess the influence of individual features on predictions. Using a prospective pediatric ophthalmology cohort of over 3000 children with longitudinal follow-up, our method estimates direct and indirect causal effects through intervention. Applied to myopia progression in children, our approach not only achieved good performance but also identified clinically plausible causal pathways. Refutation experiments with multiple falsification strategies confirm the robustness and reliability of causal effects. The approach is model-agnostic and suitable for digital health interventions requiring explainability. By incorporating unit-level causal reasoning into deep learning, this work advances transparent and reliable AI systems aligned with the goals of precision medicine and equitable healthcare.

Background: Large-scale myopia screening in the Western Pacific region often relies on noncycloplegic refraction because cycloplegia is difficult to implement routinely. However, raw noncycloplegic measurements can overestimate myopia and bias prevalence estimates used for surveillance and policy evaluation. Methods: We analyzed four Chinese cohorts spanning preschool children to young adults. Participants from Anyang and Beijing were randomly divided into derivation and internal validation sets, and the Lhasa cohort served as an external validation set. Candidate predictors included noncycloplegic spherical equivalent refraction (SER), ocular biometry, age, and sex. The primary outcome was cycloplegic spherical equivalent refraction; secondary outcomes were myopia, high myopia, and hyperopia prevalence. Findings: Among 30 352 participants, the full model explained 91.6% of the variance in cycloplegic SER. In internal validation, mean prediction difference was 0.00 ± 0.70 diopters and mean absolute error was 0.57 D. In external validation, the corresponding errors were 0.883 D and 1.088 D. The absolute error in myopia prevalence decreased from 18.29 percentage points with raw noncycloplegic refraction to 4.29 with model-based estimates in internal validation, and from 13.54 to 8.62 in external validation. This corresponded to approximately 140 and 49 fewer excess myopia classifications per 1000 screened participants, respectively. Hyperopia estimates improved similarly. Interpretation: Routine noncycloplegic screening data can support more accurate estimation of cycloplegic refractive burden than raw noncycloplegic refraction alone. This interpretable model may help improve surveillance, referral planning, and policy evaluation when cycloplegia is impractical, while cycloplegic refraction remains necessary for individual diagnosis.