Text-to-image (T2I) artificial intelligence models are being increasingly explored in medical education, yet their utility in ophthalmology remains unclear. Slit-lamp anterior segment photography, as a cornerstone of ophthalmic training, provides an ideal context for evaluation. We assessed 40 cases of anterior segment disease. The text descriptions were generated using GPT-4o, and the corresponding images were synthesized via Sora Turbo. Readability was analysed with the Flesch Reading Ease (FRE), Flesch‒Kincaid Grade Level (FKGL), and Gunning Fog Scale (GFS). Twenty ophthalmologists (10 juniors, 10 seniors) rated image-text pairs across five dimensions—text accuracy, image reliability, recognizability, educational value, and generation stability—using a 5-point Likert scale. Entities with distinct morphological features, such as cataracts and subconjunctival haemorrhages, received the highest total scores, whereas those with entropion and corneal foreign bodies scored the lowest. Readability analysis indicated advanced text complexity. Senior ophthalmologists consistently provided lower ratings than junior clinicians did, highlighting expertise-related differences in perceived educational value. Sora Turbo can generate clinically useful anterior segment images for educational purposes, particularly for pathologies with prominent morphological features. This first systematic evaluation in ophthalmology demonstrates the promise of AI-generated atlases as scalable teaching resources for early-stage trainees while emphasizing the need for expert validation and ethical oversight before integration into formal curricula.

Purpose: To compare the postoperative rotational stability of three toric intraocular lens (IOL) platforms and identify platform-specific biometric factors associated with rotation. Setting: Three eye centers in China. Design: Prospective multicenter randomized controlled trial. Methods: A total of 256 eyes of 172 cataract patients with regular corneal astigmatism that completed 3 months of follow-up after phacoemulsification with implantation of TECNIS Toric II, Clareon Toric, or AT TORBI 709M were included. Participants were randomly assigned to 1 of the 3 toric IOL platforms. Baseline rotational alignment was defined at the end of surgery and compared with retroillumination images obtained at 1 day, 3 days, 1 week, 2 weeks, 1 month, and 3 months postoperatively. Results: Absolute rotation changed most prominently within the first week and then progressively stabilized. Rotational stability was reached earlier in the ZCU and CNW0T groups, which plateaued at 1 week, whereas the 709 approached stability between 2 weeks and 1 month. Throughout follow-up, the 709 showed greater rotation than the CNW0T, whereas the ZCU demonstrated the least rotation. Significant rotation of ≥10° occurred in 1.2%, 2.4%, and 12.2% of eyes, respectively. White-to-white distance was associated with greater rotation in the CNW0T and 709 groups, lens thickness was additionally associated with greater rotation in the 709 group, and no biometric factor reached significance in the ZCU group. Conclusions: Hydrophobic C-loop toric IOLs showed better early rotational stability than the hydrophilic plate-haptic toric IOL. Biometric correlates of rotation differed across platforms, supporting individualized toric IOL selection and postoperative monitoring.

Purpose: To identify independent risk factors associated with significant postoperative rotation (≥10 degrees) of plate-haptic toric intraocular lenses (IOLs) and to develop and externally validate a predictive nomogram. Setting: 2 independent eye centers in China. Design: Prospective observational study with a training cohort and an external validation cohort. Methods: 805 eyes from 805 cataract patients with regular corneal astigmatism (≥0.75 diopter) who underwent phacoemulsification and plate-haptic toric IOL implantation between August 2021 and December 2024 were included. Patients from 1 center formed the training cohort, while patients from the second center served as the external validation cohort. Least absolute shrinkage and selection operator regression followed by multivariate logistic regression was used to identify predictors of significant toric IOL rotation (≥10 degrees at 2 weeks postoperatively). A predictive nomogram was developed and validated through receiver operating characteristic analysis, calibration curve, and decision curve analysis (DCA). Results: Age (odds ratio [OR] = 1.04, P = .002), anterior chamber depth (ACD) (OR = 7.71, P < .001), lens thickness (LT) (OR = 5.13, P < .001), and white-to-white (WTW) (OR = 2.36, P = .002) were identified as independent predictors. The nomogram demonstrated good discriminative performance with an area under the curve of 0.809 in the training cohort and 0.848 in the validation cohort. Calibration and DCA analyses confirmed the accuracy and clinical utility of the model. Conclusions: A validated nomogram based on age, ACD, LT, and WTW provides a useful tool for individualized preoperative risk assessment of significant toric IOL rotation, aiding surgical decision-making.

Purpose: To evaluate the effect of dry eye disease (DED) and ocular surface treatment on the agreement and repeatability of corneal astigmatism measurements obtained using IOLMaster 700, Pentacam, and iTrace in patients with age-related cataract. Setting: Eye, Ear, Nose, and Throat Hospital of Fudan University, Shanghai, China. Design: Prospective, single-center study. Methods: A total of 313 eyes of patients with age-related cataract were included. Eyes with DED received 0.3% sodium hyaluronate eye drops for 2 weeks and were re-examined after treatment. Corneal astigmatism measurements were obtained using IOLMaster 700, Pentacam, and iTrace before pupil dilation. The main parameters included keratometric corneal astigmatism, total corneal astigmatism, and their vector components. Interdevice agreement was assessed using Bland-Altman analysis, and test-retest repeatability was evaluated using within-subject standard deviation, test-retest repeatability, and intraclass correlation coefficients. Results: Compared with the control group, eyes at DED baseline showed wider limits of agreement and poorer TRT for both scalar astigmatism parameters and vector components. After ocular surface treatment, LoA width decreased by 50.5%–62.5% for KCA and 19.9%–29.0% for TCA across the three device comparisons. KCA repeatability improved significantly after treatment for IOLMaster 700 and iTrace, whereas significant improvement in TCA repeatability was observed only for IOLMaster 700. Device- and parameter-specific differences persisted after treatment. Conclusions: DED was associated with reduced agreement and repeatability of corneal astigmatism measurements obtained using IOLMaster 700, Pentacam, and iTrace. Short-term ocular surface treatment was associated with improved measurement stability.

Background: Contemporary multimodal large language models (LLMs) can process both clinical text and medical images, but their reliability in retinal imaging-based question answering remains uncertain. This study evaluated the performance of six contemporary multimodal LLMs on retinal multiple-choice questions derived from OCTCases. Methods: This cross-sectional benchmark study included 226 multiple-choice questions from 78 OCTCases retinal cases, comprising 151 image-based and 75 nonimage-based questions. Six multimodal LLMs were evaluated: ChatGPT-5.5 Instant, ChatGPT-5.5 Thinking, Grok 4, Gemini 3, DeepSeek V4, and Kimi 2.5. Model-selected answers were compared with the OCTCases answer key, which was reviewed by three retina specialists. Accuracy was assessed overall and by question subset. Pairwise model comparisons, inter-model agreement, voting-based group-answer performance, item difficulty distribution, and interface latency were analyzed. Results: Overall accuracy ranged from 59.3 to 77.4%, with the highest accuracy observed for ChatGPT-5.5 Thinking, followed by Gemini 3 and ChatGPT-5.5 Instant. DeepSeek V4 showed the lowest overall accuracy. All models performed better on nonimage-based questions than on image-based questions. In the image-based subset, Gemini 3 achieved the highest accuracy, whereas DeepSeek V4 showed the lowest accuracy. Under the majority-vote rule, group answers were generated for 182 of 226 questions and achieved an accuracy of 89.6% among questions with a determinate group answer. The plurality-vote rule generated group answers for more questions but with lower accuracy. Inter-model agreement was generally higher for nonimage-based than image-based questions, and all questions answered incorrectly by all six models were image-based. Interface latency varied across models and showed no consistent association with response correctness. Conclusion: Contemporary multimodal LLMs showed promising but uneven performance on OCTCases retinal questions. Their performance was consistently better on nonimage-based than image-based questions, indicating that retinal image interpretation remains a major limitation. Voting-based group answers may provide a useful reliability signal when model agreement is present, whereas model disagreement may help identify difficult or visually ambiguous questions requiring specialist review. These findings support continued benchmarking and cautious, specialist-supervised use of multimodal LLMs in retinal education and structured imaging question answering.