Chaoyu Lei

All Publications

• ProptoView: AI-based digital exophthalmometry using multi-view facial images in a multinational validation study. Journal of translational medicine Lei, C., Song, S., Fan, J., Pandiyan, P. S., Ding, J., Wattanaphanich, S., Liu, X., Lu, W., Wei, D., Zhang, S., Zhou, M., Su, J., Lyu, X., Zhuang, W., Song, X., Xu, B., Ding, X., Sintuwong, S., Yip, C. C., Dang, K., Zhou, H. 2026; 24 (1)

  Abstract

  Accurate proptosis measurement is vital for managing thyroid eye disease (TED) and other orbital conditions. However, current approaches have certain limitations: the Hertel exophthalmometer is convenient but imprecise, while computed tomography (CT) is accurate but costly and exposes patients to radiation.We developed ProptoView, an AI-based digital exophthalmometer, using 5676 images from 2516 eyes across 1258 visits of 763 TED patients with CT and Hertel measurements. For external validation, we used an additional 644 images from 648 eyes of 324 patients with TED and other orbital diseases, collected across three countries and five hospitals. Patients provided up to five images from four views. A three-stage deep learning approach, optimized with Adam and validated via five-fold cross-validation, helped develop three AI models: single-view, multi-view, and dynamic input.Compared with CT, the single-view model achieved an intraclass correlation coefficient (ICC) of 0.859, slightly lower than the Hertel exophthalmometer's ICC of 0.888. The multi-view model achieved an ICC of 0.890, surpassing the Hertel exophthalmometer (0.871). The dynamic input model achieved the highest accuracy with an ICC of 0.901. Among two-view combinations, pairing the frontal view with another angle showed the highest agreement when paired with the upward gaze view (ICC = 0.855). In external validation, ProptoView showed robust concordance with the Hertel exophthalmometer (ICC = 0.845), comparable to its agreement in the development dataset. Additionally, ProptoView reduced misclassification at the 19-mm threshold (14.7% vs. 20.5% with the Hertel exophthalmometer).ProptoView provides an accurate, non-contact, and cost-effective solution for proptosis measurement. Its flexibility and precision suggest significant potential for streamlining clinical workflows and enabling telemedicine applications.

  View details for DOI 10.1186/s12967-026-08401-w

  View details for PubMedID 42271356

  View details for PubMedCentralID PMC13330175

• Centering the marginalized: AI-driven strategies for advancing health equity in rare disease care. Patterns (New York, N.Y.) Lei, C., Zuo, Y., Lopes, C. A., Stuart, J., Xu, B., Liu, T. Y., Ploug, T., Wang, Z., Dang, K., Jin, K., Yu, H., Tatere, H. Y., Kong, F., Zhang, N., Zhang, L., Zhou, H. 2026; 7 (5): 101535

  Abstract

  Rare diseases (RDs) affect 6%-8% of the global population but remain critically underserved. People living with an RD face misdiagnosis, limited treatment options, and inequitable access to specialized care. While artificial intelligence (AI) offers transformative potential in RD care, significant challenges remain. This perspective identifies five key dimensions to equitable AI application in RD care: data availability, algorithmic fairness, patient privacy, resource prioritization, and medical ethics. To address these barriers, strategies include enhancing data diversity through internationally harmonized repositories, leveraging synthetic data, and employing fairness-aware algorithms. Privacy-preserving methods safeguard sensitive genetic data while enabling collaborative research. Transparent resource-allocation frameworks and interdisciplinary governance ensure equitable distribution of AI-driven benefits, particularly in low- and middle-income countries. Ethical considerations, including patient-centered consent and dynamic risk assessments, are foundational to sustainable AI integration. By addressing these multidisciplinary challenges, AI can advance health equity, transforming RD care from fragmented and inequitable to inclusive and innovative. This paradigm shift aligns technological progress with the ethical imperative to ensure no patient is left behind in the promise of precision medicine.

  View details for DOI 10.1016/j.patter.2026.101535

  View details for PubMedID 42130943

  View details for PubMedCentralID PMC13161689

• Benchmarking clinical knowledge and multi-modal reasoning of large language models in liver cirrhosis SCIENTIFIC REPORTS Zhang, Y., Chen, H., Xu, R., Lei, C., Han, D., Li, Y., Wang, X. 2026; 16 (1)

  Abstract

  As large language models (LLMs) become increasingly integral to healthcare, patients are frequently turning to them in the long-term management of chronic conditions like liver cirrhosis. However, the lack of standardized benchmarks complicates the selection of the most suitable models for specific clinical tasks. To address this gap, we constructed a question bank comprising 462 multiple-choice, 25 short-answer, and 40 multi-modal case questions to evaluate six mainstream LLMs in terms of clinical knowledge and multi-modal reasoning capability related to liver cirrhosis. Gemini-2.5pro dominated structured knowledge tasks with 88.5% accuracy in multiple-choice questions, while Grok-4 excelled in multi-modal reasoning, achieving 86.7% accuracy in case questions and outperforming Gemini-2.5pro across all dimensions in specialist-evaluated short-answer responses. Task-specific analysis further revealed complementary strengths, with GPT-5 excelling in diagnosis and DeepSeek-R1 in test interpretation. These findings highlight the distinct advantages of different models, underscoring the potential of LLMs as valuable auxiliary tools for medical education and decision support in cirrhosis management.

  View details for DOI 10.1038/s41598-026-44685-0

  View details for Web of Science ID 001795904600014

  View details for PubMedID 42020640

  View details for PubMedCentralID PMC13273182

• Evaluating and enhancing the performance of large language models in thyroid eye disease through customization and Chain-of-Thought strategies. Scientific reports Lei, C., Zhao, C., Ji, K., Cao, C., Chen, H., Wei, D., Zhang, S., Guo, S., Li, R., Zhong, S., Song, X., Zhai, G., Hoi, M., Zheng, Z., Guan, H., Zhou, H. 2026; 16 (1)

  Abstract

  The paucity of public awareness regarding thyroid eye disease (TED) usually leads to delayed medical care. While large language models (LLMs) hold great potential for augmenting patient education, their ability in answering TED-related questions has yet to be comprehensively evaluated. This study aims to assess the capability of LLMs to address TED-related questions and explore the practicability of customizing LLMs for disease-specific domains. Considering the diverse LLM candidates, we deployed a cascade pipeline to search for the best model for TED. We first evaluated performances of several prevailing LLMs on multiple-choice questions. The best-performing models, GPT-4 and Claude 3.5, were selected and customized to create TED-GPT and TED-Claude. Chain-of-Thought (CoT) was then utilized, resulting in CoT-GPT and CoT-Claude. We also evaluated newer LLMs with native CoT capabilities (GPT-4-o1, GPT-4-o3, Gemini-2.0-Flash, Gemini-2.5-Pro, Claude 3.7). These models, along with their original versions, were then assessed and compared on multiple-choice questions. The better-performing TED-GPT and TED-Claude were evaluated on short-answer and case questions, with comparisons made to their original ones using the QUEST framework (Quality, Understanding/Reasoning, Expression, Safety/Harm, Trust). For multiple-choice questions, GPT-4 and Claude 3.5 arrived competitive accuracies (76.2% and 83.2%, respectively). The addition of CoT, along with customization into GPT-4 and Claude 3.5, led to improvement in the accuracy (CoT-GPT 86.1%, CoT-Claude 87.1%, TED-GPT 86.1%, TED-Claude 89.1%), outperforming all other newer LLMs. For case and short-answer questions, the customized TED-GPT and TED-Claude also performed better than their original versions. TED-Claude showed the best performance in accuracy, readability, comprehensiveness, likelihood of harm and reasoning. Thus, LLMs, particularly TED-Claude, achieved relatively satisfactory performances in answering TED-related questions. Besides, using LLMs’ customization modules, along with CoT, effectively enhanced model performances. This indicates that clinicians can use these simple and universal methods to construct LLMs suitable for specific medical domains.

  View details for DOI 10.1038/s41598-026-48996-0

  View details for PubMedID 42000844

  View details for PubMedCentralID PMC13254332

• Camurati-Engelmann disease with bilateral proptosis and optic neuropathy: a case report and literature review. Orbit (Amsterdam, Netherlands) Lei, C., He, Z., Liang, Y., Li, Y., Zhao, S. X., Qiao, J., Zhou, H. 2026: 1-8

  Abstract

  Camurati-Engelmann disease (CED) is a rare autosomal dominant skeletal disorder caused by mutations in the TGFB1 gene and is characterized by progressive diaphyseal widening and cortical thickening of long bones. A 29-year-old woman presented with bilateral proptosis, hypothyroidism and a 15-year history of hearing loss and limb pain. Superior visual field defects were detected on automated perimetry. Orbital computed tomography demonstrated orbital apex crowding and narrowing of the optic canals caused by skull hyperostosis, raising concern for compressive optic neuropathy. Imaging revealed systemic skeletal abnormalities, including macrocephaly, mandibular overgrowth, and cortical thickening of multiple bones. Genetic testing confirmed a pathogenic TGFB1 mutation, establishing the diagnosis of CED. Treatment with zoledronic acid, methotrexate, and hormone replacement therapy improved bone metabolism and relieved bone pain at the six-month follow-up. At one-year follow-up, optical coherence tomography showed reduced peripapillary retinal nerve fiber layer thickness (mean 66 μm in both eyes), accompanied by mild progression of visual field defects. Orbital apex decompression with possible optic canal decompression was considered if visual function declined further. This case highlights the potential for orbital involvement and compressive optic neuropathy in CED and underscores the importance of multidisciplinary management and regular ophthalmologic monitoring.

  View details for DOI 10.1080/01676830.2026.2650802

  View details for PubMedID 41949641

• Management of Thyroid Eye Disease: A Comparison Between Three Recent Clinical Guidelines. Ophthalmology and therapy Lei, C., Jin, J., Li, R., Yip, C. C., Chong, K. K., Smith, T. J., Sun, J., Zhou, H. 2026; 15 (3): 985-999

  Abstract

  Rapid advances in the diagnosis and treatment of thyroid eye disease (TED) have led to the development of three patient care guidelines by regional professional societies: the European Group on Graves' Orbitopathy, the Oculoplastics and Orbital Diseases Group of the Chinese Medical Association Ophthalmology Branch/Thyroid Group of the Chinese Medical Association Endocrinology Branch, and the American Thyroid Association/European Thyroid Association. Although broad consensus can be found across the three guidelines, important differences could affect patient management. This review examines and compares the recommendations of these guidelines across 11 dimensions, from disease diagnosis to treatment strategies. We explore the possible root sources of these variations. The review also suggests future directions and potential implications, thus providing a comprehensive perspective of current and future management of TED.

  View details for DOI 10.1007/s40123-026-01326-z

  View details for PubMedID 41697453

  View details for PubMedCentralID PMC12976230

• Sequential sensitivity analysis of multimodal large language models for rare orbital disease detection. Communications medicine Lei, C., Ji, K., Zhao, C., Zhong, S., Cao, C., Chen, H., Yip, C. C., Sintuwong, S., Ding, J., Pandiyan, P. S., Wattanaphanich, S., Johnston, L., Ren, Y., Duan, X., Zhai, G., Zhou, H. 2026; 6 (1)

  Abstract

  Delayed diagnosis of rare orbital diseases is attributed to limited clinical awareness. Building on prior evidence of multimodal large language models (MLLMs) for detecting common ocular conditions, this study aims to evaluate whether integrating multimodal clinical data can enhance the diagnostic accuracy of MLLMs for rare orbital diseases.We conducted a multinational, multiracial, retrospective study. Two datasets were analyzed: Dataset 1, containing 6,786 single-eye photographs from China, was used to fine-tune a contrastive language-image pre-training (CLIP) for preliminary classification of healthy eyes, orbital diseases, and non-orbital diseases, and to compare its performance against three traditional models and three next-generation models. Dataset 2, comprising 170 participants from China, Singapore, and Thailand, was used to evaluate a MLLM (GPT-4o-Latest). Sequential sensitivity analysis assessed the impact of adding external eye photographs, chief complaints, racial information, and diagnostic reasoning prompts. An AI agent combining the CLIP model with GPT-4o-Latest was further evaluated. The model's ability to generate medical reports and examination recommendations was also assessed.Here we show that the CLIP model achieves 90.21% preliminary detection accuracy, surpassing all baseline models. MLLM detection accuracy improves significantly with the inclusion of multimodal inputs. When relying on external eye images, the top-5 accuracy is 25.68%. The combined agent raises top-5 accuracy to 85.29%. Generated reports and recommendations display high accuracy, readability, completeness, and low potential for harm.Our study demonstrates the potential of MLLM in improving diagnostic accuracy and supporting clinical decision-making for rare orbital diseases.

  View details for DOI 10.1038/s43856-026-01447-3

  View details for PubMedID 41720932

  View details for PubMedCentralID PMC13035938

• Enhancing ocular sign detection: AI-based strategic segmentation for improved accuracy and privacy protection. NPJ digital medicine Lei, C., Zhao, C., Chen, J., Duan, X., Xu, C., Yip, C. C., Sintuwong, S., Ding, J., Pandiyan, P. S., Wattanaphanich, S., Ren, Y., Luo, S., Song, X., He, H., Liu, X., Zhou, H. 2026; 9 (1): 130

  Abstract

  Accurate detection of ocular signs is essential for early diagnosis of eye diseases, but current AI approaches using facial or external ocular images include non-essential information, compromising performance and patient privacy. We conducted a multinational retrospective study of 2360 eyes from 1180 half-face images of thyroid eye disease patients across five racial groups from five hospitals in three countries. We developed a Dense Squeeze-and-Excitation Network (DSE-Net) to segment eyelid, conjunctiva, lacrimal caruncle, and eyeball, minimizing exposure and enhancing privacy. DSE-Net achieved Dice coefficient of 84.7%, 84.8%, 92.7%, and 95.1%, outperforming seven segmentation models. We then built SegmenView, employing LeNet, AlexNet, ResNet50, and VGGNet16 to detect eyelid edema, conjunctival erythema, caruncle or plica edema, and exophthalmos. SegmenView achieved internal Area Under the Curve (AUCs) of 71.09%, 80.81%, 90.07%, and 82.86%; external AUCs ranging 55.58%-84.29% across two test datasets, outperforming half-face and periocular models. We also compared SegmenView with four privacy-preserving methods, showing its superior ability to balance privacy protection with diagnostic accuracy. Additionally, visualizations based on Gradient-weighted Class Activation Mapping (Grad-CAM) further enhanced the model's interpretability. Our approach demonstrates high accuracy, generalizability, and potential for lightweight, privacy-preserving ocular sign detection.

  View details for DOI 10.1038/s41746-025-02310-w

  View details for PubMedID 41495351

  View details for PubMedCentralID PMC12881351

• Quantitative analysis of conjunctival vascular alterations: Applications in ocular and systemic disease detection. Progress in retinal and eye research Duan, X., Lei, C., Lim, C. H., Ding, J., Mehta, J. S., Basu, S., Johnston, L., Ren, Y., Zhao, C., Chang, V. K., Zhou, H. 2026; 110: 101416

  Abstract

  The conjunctival blood vessels are the only microcirculatory system on the body surface that can be observed non-invasively. Anatomically interconnected with multiple craniofacial circulatory systems, these vessels can indirectly reflect the blood supply to these areas and the overall state of systemic microcirculation. We synthesize findings from 48 studies spanning 2020-2025. Overall, existing research has found that the conjunctival vascular parameters can change in various diseases such as diabetes, cardiovascular diseases, and autoimmune diseases, and may even precede organic lesions. Recent advancements in conjunctival vessel imaging and analysis technologies have enabled the identification and evaluation of various ocular and systemic diseases based on conjunctival vascular parameters. However, existing studies are limited by insufficient sample sizes, covariate interference, limited disease types, a lack of investigation into the underlying mechanisms of conjunctival vascular changes, and inadequate integration with emerging technologies, such as artificial intelligence. Future research should aim to broaden the scope of investigation, delve deeper into the mechanisms governing conjunctival vascular alterations, and integrate artificial intelligence to establish a solid foundation for the clinical application of conjunctival vascular parameters.

  View details for DOI 10.1016/j.preteyeres.2025.101416

  View details for PubMedID 41260358

• ROFI: a deep learning-based ophthalmic sign-preserving and reversible patient face anonymizer. NPJ digital medicine Tian, Y., Zhou, M., Chen, Y., Li, F., Qi, L., Wang, S., Xu, X., Yu, Y., Xu, S., Lei, C., Jiang, Y., Zhang, R., Tan, J., Wu, L., Chen, H., Liu, X., Lu, W., Li, L., Zhou, H., Song, X., Zhai, G., Fan, X. 2025; 8 (1): 705

  Abstract

  Patient face images provide a convenient mean for evaluating eye diseases, while also raising privacy concerns. Here, we introduce ROFI, a deep learning-based privacy protection framework for ophthalmology. Using weakly supervised learning and neural identity translation, ROFI anonymizes facial features while retaining disease features (over 98% accuracy, κ > 0.90). It achieves 100% diagnostic sensitivity and high agreement (κ > 0.90) across eleven eye diseases in three cohorts, anonymizing over 95% of images. ROFI works with AI systems, maintaining original diagnoses (κ > 0.80), and supports secure image reversal (over 98% similarity), enabling audits and long-term care. These results show ROFI's effectiveness of protecting patient privacy in the digital medicine era.

  View details for DOI 10.1038/s41746-025-02062-7

  View details for PubMedID 41266592

  View details for PubMedCentralID PMC12635282

• Clinical Features of Chinese Patients with Thyroid Eye Disease: A Multicenter Retrospective Study. Thyroid : official journal of the American Thyroid Association Lei, C., Lyu, X., Ren, Y., Wei, D., Zhang, S., Zhang, Y., Wang, L., Liu, L., Wen, J., Liu, X., Lin, C., Lu, W., Liu, J., Li, R., Zhang, S., Song, X., Yu, Z., Bahn, R. S., Zhou, H. 2025; 35 (10): 1187-1197

  Abstract

  Background: Thyroid eye disease (TED) is a debilitating autoimmune disorder linked to thyroid dysfunction. There is limited knowledge of TED in Asian populations. This multicenter study characterizes the clinical features and treatment response of TED in a large Chinese cohort. Methods: A retrospective multicenter study included 4157 patients with TED from nine Chinese hospitals from February 2016 to July 2023. Disease severity and activity were evaluated according to the European Group on Graves' Orbitopathy standards. We examined associations of variables including sex, age, smoking status, I131 treatment, consultation department, and geographical region with clinical outcomes. Logistic regression and nomogram models were developed to examine associations with sight-threatening TED and, in a subgroup analysis (n = 126), patients' responsiveness to intravenous glucocorticoid (IVGC) therapy. Results: We included 4157 patients with mean age and standard deviation (SD) 45.96 ± 16.44 years. Of these, 63.6% (n = 2644) were females. Over half (55.6%, n = 2310) of participants were in the inactive phase, with a mean clinical activity score of 2.19 ± 1.61 (SD) for all patients. TED severity was categorized as mild (9.3%, n = 385), moderate-to-severe (82.5%, n = 3428), and sight-threatening (8.2%, n = 344). The average degree of exophthalmos was 20.04 ± 5.27 mm, and 48.8% (n = 2029) of patients had diplopia. Patients treated with I131 had higher disease activity (47.5%, n = 468, vs. 43.5%, n = 1379, p < 0.05). Coastal region patients exhibited more severe TED (sight-threatening cases: 10.1%, n = 195, vs. 7.2%, n = 147) and higher diplopia scores (1.00 ± 1.10 vs. 0.86 ± 1.09, p < 0.001) than inland counterparts. TED severity was also greater in patients treated in Ophthalmology departments (mild cases: 6.0%, n = 213; moderate-to-severe cases: 85.6%, n = 3055) compared with Endocrinology departments (mild cases: 29.3%, n = 172; moderate-to-severe cases: 63.5%, n = 373). Nomograms had an area under the receiver operating curve of 0.742 (confidence interval [CI] 0.716-0.768) for sight-threatening TED and 0.759 (CI 0.674-0.843) for IVGC therapy responsiveness. Conclusions: We characterized the clinical features and treatment response of TED in a large Chinese cohort. These findings offer valuable insights informing TED risk stratification in Asian patients and forming a foundation for future prospective studies.

  View details for DOI 10.1177/10507256251359559

  View details for PubMedID 40658134

• Integrating ocular and clinical features to enhance intravenous glucocorticoid response prediction in thyroid eye disease: a machine learning approach. Endocrine Zhao, C., Lei, C., Pei, S., Ren, Y., Duan, X., Guo, S., Song, X., Wang, H., Zhou, H. 2025; 90 (1): 188-198

  Abstract

  Predicting intravenous glucocorticoid (IVGC) efficacy in thyroid eye disease (TED) is vital for personalized treatment and minimizing side effects. Current methods haven't fully utilized ocular features. This study aims to integrate ocular features into predictive model to assess their impact on improving IVGC efficacy prediction.This retrospective study recruited 130 TED patients who received 4.5 g of IVGC treatment and collected their clinical features. After Least Absolute Shrinkage and Selection Operator (LASSO) regression for feature selection, two key features, lid aperture and CAS, were identified and incorporated into a predictive model. Subsequently, five ocular features were added, resulting in a model using both clinical and ocular features. Six machine learning classifiers were tested on both models, and the performances of two models were compared. The best-performing predictive model was analyzed using SHapley Additive exPlanations (SHAP) to interpret the model.In the LASSO regression, CAS and lid aperture were selected as key features for predicting IVGC efficacy. In the model using only clinical features, the best-performing classifier was Logistic Regression, with an AUC of 0.701. However, when ocular features were incorporated, the XGBoost classifier outperformed all others, with the AUC improving to 0.821. SHAP analysis further indicated that conjunctival edema was the most important feature for prediction.This study identified features associated with the prediction of IVGC efficacy and demonstrated that incorporating ocular features into clinical parameters improves the ability to predict treatment outcomes. Additionally, SHAP analysis highlighted the importance of ocular features in predicting treatment efficacy, providing a basis for further mechanistic exploration.

  View details for DOI 10.1007/s12020-025-04300-0

  View details for PubMedID 40549141

  View details for PubMedCentralID 9723260

• Penetrating orbital injury by a nine-centimetre lead sinker. BMJ case reports Lei, C., Li, Y., Zhou, H., Zhong, S. 2025; 18 (7)

  Abstract

  A male in his early 40s presented with a penetrating orbital injury after a nine-centimetre lead sinker was propelled into his right orbit while fishing. He reported pain, ptosis and visual acuity reduced to light perception. X-rays, preferred over CT due to metallic artefacts, revealed the sinker in the inferior orbital fissure with an intact eyeball. Initial surgical extraction attempts triggered a severe vagal response, necessitating endoscopic navigation for safe removal. Postoperatively, despite an intact globe, he developed vitreous haemorrhage and retinal detachment, requiring vitrectomy and silicone oil injection. Three months later, his visual acuity improved to 20/200, with normal blood lead levels. This case emphasises selecting imaging based on foreign body material, avoiding blind extraction using advanced tools if needed, monitoring for intraocular complications and assessing systemic toxicity risks.

  View details for DOI 10.1136/bcr-2025-265477

  View details for PubMedID 40628683

  View details for PubMedCentralID PMC12243092

• A novel spatial-temporal image fusion method for augmented reality-based endoscopic surgery. Medical image analysis Shi, H., Xu, J., Li, H., Jiang, S., Lei, C., Zhou, H., Li, Y., Chen, X. 2025; 103: 103609

  Abstract

  Augmented reality (AR) has significant potential to enhance the identification of critical locations during endoscopic surgeries, where accurate endoscope calibration is essential for ensuring the quality of augmented images. In optical-based surgical navigation systems, asynchrony between the optical tracker and the endoscope can cause the augmented scene to diverge from reality during rapid movements, potentially misleading the surgeon-a challenge that remains unresolved. In this paper, we propose a novel spatial-temporal endoscope calibration method that simultaneously determines the spatial transformation from the image to the optical marker and the temporal latency between the tracking and image acquisition systems. To estimate temporal latency, we utilize a Monte Carlo method to estimate the intrinsic parameters of the endoscope's imaging system, leveraging a dataset of thousands of calibration samples. This dataset is larger than those typically employed in conventional camera calibration routines, rendering traditional algorithms computationally infeasible within a reasonable timeframe. By introducing latency as an independent variable into the principal equation of hand-eye calibration, we developed a weighted algorithm to iteratively solve the equation. This approach eliminates the need for a fixture to stabilize the endoscope during calibration, allowing for quicker calibration through handheld flexible movement. Experimental results demonstrate that our method achieves an average 2D error of 7±3 pixels and a pseudo-3D error of 1.2±0.4mm for stable scenes within 82.4±16.6 seconds-approximately 68% faster in operation time than conventional methods. In dynamic scenes, our method compensates for the virtual-to-reality latency of 11±2ms, which is shorter than a single frame interval and 5.7 times shorter than the uncompensated conventional method. Finally, we successfully integrated the proposed method into our surgical navigation system and validated its feasibility in clinical trials for transnasal optic canal decompression surgery. Our method has the potential to improve the safety and efficacy of endoscopic surgeries, leading to better patient outcomes.

  View details for DOI 10.1016/j.media.2025.103609

  View details for PubMedID 40334584

• Reshaping the hierarchical medical system for rare diseases: a two-tier structure and one-stop referral network. Journal of global health Lei, C., Zuo, Y., Kong, F., Chen, H., Cheng, H., Song, X., Zhang, L., Zhou, H. 2025; 15: 03005

  View details for DOI 10.7189/jogh.15.03005

  View details for PubMedID 40208802

  View details for PubMedCentralID PMC11984612

• AI-assisted facial analysis in healthcare: From disease detection to comprehensive management. Patterns (New York, N.Y.) Lei, C., Dang, K., Song, S., Wang, Z., Chew, S. P., Bian, R., Yang, X., Guan, Z., Lopes, C. l., Wang, M. H., Choy, R. W., Hu, X., Lai, K. K., Chong, K. K., Pang, C. P., Song, X., Su, J., Ding, X., Zhou, H. 2025; 6 (2): 101175

  Abstract

  Medical conditions and systemic diseases often manifest as distinct facial characteristics, making identification of these unique features crucial for disease screening. However, detecting diseases using facial photography remains challenging because of the wide variability in human facial features and disease conditions. The integration of artificial intelligence (AI) into facial analysis represents a promising frontier offering a user-friendly, non-invasive, and cost-effective screening approach. This review explores the potential of AI-assisted facial analysis for identifying subtle facial phenotypes indicative of health disorders. First, we outline the technological framework essential for effective implementation in healthcare settings. Subsequently, we focus on the role of AI-assisted facial analysis in disease screening. We further expand our examination to include applications in health monitoring, support of treatment decision-making, and disease follow-up, thereby contributing to comprehensive disease management. Despite its promise, the adoption of this technology faces several challenges, including privacy concerns, model accuracy, issues with model interpretability, biases in AI algorithms, and adherence to regulatory standards. Addressing these challenges is crucial to ensure fair and ethical use. By overcoming these hurdles, AI-assisted facial analysis can empower healthcare providers, improve patient care outcomes, and enhance global health.

  View details for DOI 10.1016/j.patter.2025.101175

  View details for PubMedID 40041850

  View details for PubMedCentralID PMC11873005

• Research beneficiaries speak. Science (New York, N.Y.) Ye, Q., Gao, Y., Patel, R. J., Cao, B., Li, A., Powell, J., Cai, Y., Marín-Beltrán, I., Tao, Z., Ejaz, M., Zaidi, S. S., Gupta, A., See, M. S., Cao, X. E., Fulco, G., Lei, C., Han, S., Teng, D., Bezerra, P., Xia, Y., Montesdeoca Molina, B. R., Zhang, T., Heim, A. B., Uzonyi, A. 2024; 384 (6691): 26-28

  View details for DOI 10.1126/science.adp2180

  View details for PubMedID 38574143

• Revisiting global health education: The engagement of medical students. Journal of global health Lei, C., Zhou, H., Qu, M., Cheng, H., Tao, Z., Song, X., Zhou, H., Zhang, L. 2023; 13: 03059

  View details for DOI 10.7189/jogh.13.03059

  View details for PubMedID 37917877

  View details for PubMedCentralID PMC10623375

• Facial expression of patients with Graves' orbitopathy. Journal of endocrinological investigation Lei, C., Qu, M., Sun, H., Huang, J., Huang, J., Song, X., Zhai, G., Zhou, H. 2023; 46 (10): 2055-2066

  Abstract

  Patients with Graves' orbitopathy (GO) have characteristic facial expressions that are different from those of healthy individuals due to the combination of somatic and psychiatric symptoms. However, the facial expressions of GO patients have not yet been described and analyzed systematically. Thus, the present study aimed to present the facial expressions of GO patients and explore their applications in clinical practice.Facial image and clinical data of 943 GO patients were included, and 126 patients answered quality of life (GO-QOL) questionnaires. Each patient was labeled for one facial expression. Then, a portrait was drawn for every facial expression. Logistic and linear regression was performed to analyze the correlation between facial expression and clinical indicators, including QOL, disease activity and severity. The VGG-19 network model was utilized to discriminate facial expressions automatically.Two groups, i.e., the non-negative emotion (neutral, happy) and the negative emotion (disgust, angry, fear, sadness, surprise), and seven expressions of GO patients were systematically analyzed. Facial expression was statistically associated with GO activity (P = 0.002), severity (P < 0.001), QOL visual functioning subscale scores (P = 0.001), and QOL appearance subscale score (P = 0.012). The deep learning model achieved satisfactory results (accuracy 0.851, sensitivity 0.899, precision 0.899, specificity 0.720, F1 score 0.899, and AUC 0.847).As a novel clinical sign, facial expression holds the potential to be incorporated into GO assessment system in the future. The discrimination model may assist clinicians in real-life patient care.

  View details for DOI 10.1007/s40618-023-02054-y

  View details for PubMedID 37005981

  View details for PubMedCentralID 1722683

• Layperson?s performance on an unconversant type of AED device: A prospective crossover simulation experimental study WORLD JOURNAL OF EMERGENCY MEDICINE Lei, C., Qin, H., Dong, X., You, J., Zhang, L. 2022; 13 (2): 98-105

  Abstract

  Diverse models of automated external defibrillators (AEDs) possess distinctive features. This study aimed to investigate whether laypersons trained with one type of AED could intelligently use another initial contact type of AED with varying features.This was a prospective crossover simulation experimental study conducted among college students. Subjects were randomly trained with either AED1 (AED1 group) or AED2 (AED2 group), and the AED operation performance was evaluated individually (Phase I test). At the 6-month follow-up AED performance test (Phase II test), half of the subjects were randomly switched to use another type of AED, which formed two switches (Switch A: AED1-1 group vs. AED2-1 group; Switch B: AED2-2 group vs. AED1-2 group).A total of 224 college students participated in the study. In the phase I test, a significantly higher proportion of successful defibrillation and shorter shock delivery time to achieve successful defibrillation was observed in the AED2 group than in the AED1 group. In the phase II test, no statistical differences were observed in the proportion of successful defibrillation in Switch A (51.4% vs. 36.6%, P=0.19) and Switch B (78.0% vs. 53.7%, P=0.08). The median shock delivery time within participants achieving successful defibrillation was significantly longer in the switched group than that of the initial group in both Switch A (89 [81-107] s vs. 124 [95-135] s, P=0.006) and Switch B (68 [61.5-81.5] s vs. 95.5 [55-131] s, P<0.001).College students were able to effectively use AEDs different from those used in the initial training after six months, although the time to shock delivery was prolonged.

  View details for DOI 10.5847/wjem.j.1920-8642.2022.024

  View details for Web of Science ID 000804607900003

  View details for PubMedID 35237362

  View details for PubMedCentralID PMC8861341

• A clinical decision model based on machine learning for ptosis. BMC ophthalmology Song, X., Tong, W., Lei, C., Huang, J., Fan, X., Zhai, G., Zhou, H. 2021; 21 (1): 169

  Abstract

  To establish a decision model based on two- (2D) and three-dimensional (3D) eye data of patients with ptosis for developing personalized surgery plans.Data of this retrospective, case-control study was collected from March 2019 to June 2019 at the Department of Ophthalmology, Shanghai Ninth People's Hospital, and then the patients were followed up for 3 months. One hundred fifty-two complete feature eyes from 100 voluntary patients with ptosis and satisfactory surgical results were selected, with 48 eyes excluded due to any severe condition or improper collection and shooting angle. Three experimental schemes were set as follows: use 2D distance alone, use 3D distance alone, and use two distances at the same time. The five most common evaluation indicators used in the binary classification problem to test the decision model were accuracy (ACC), precision, recall, F1-score, and area under the curve (AUC).For diagnostic discrimination, recall of "3D", "2D" and "Both" schemes were 0.875, 0.875 and 0.938 respectively. And precision of the three schemes were 0.8333, 0.7778 and 1.0000 for the surgical procedure classification. Values of "Both" scheme that combined 2D and 3D data were the highest in two classifications.In this study, 3D eye data are introduced into clinical practice to construct a decision model for ptosis surgery. Our decision model presents exceptional prediction effect, especially when 2D and 3D data employed jointly.

  View details for DOI 10.1186/s12886-021-01923-5

  View details for PubMedID 33836706

  View details for PubMedCentralID PMC8033720

• Mechanistic insights into the effect of phosphorylation on Ras conformational dynamics and its interactions with cell signaling proteins COMPUTATIONAL AND STRUCTURAL BIOTECHNOLOGY JOURNAL Wang, Y., Ji, D., Lei, C., Chen, Y., Qiu, Y., Li, X., Li, M., Ni, D., Pu, J., Zhang, J., Fu, Q., Liu, Y., Lu, S. 2021; 19: 1184-1199

  Abstract

  Ras undergoes interconversion between the active GTP-bound state and the inactive GDP-bound state. This GTPase cycle, which controls the activities of Ras, is accelerated by Ras GTPase-activating proteins (GAPs) and guanine nucleotide exchange factors (SOS). Oncogenic Ras mutations could affect the GTPase cycle and impair Ras functions. Additionally, Src-induced K-Ras Y32/64 dual phosphorylation has been reported to disrupt GTPase cycle and hinder Ras downstream signaling. However, the underlying mechanisms remain unclear. To address this, we performed molecular dynamics simulations (~30 μs in total) on unphosphorylated and phosphorylated K-Ras4B in GTP- and GDP-bound states, and on their complexes with GTPase cycle regulators (GAP and SOS) and the effector protein Raf. We found that K-Ras4B dual phosphorylation mainly alters the conformation at the nucleotide binding site and creates disorder at the catalytic site, resulting in the enlargement of GDP binding pocket and the retard of Ras-GTP intrinsic hydrolysis. We observed phosphorylation-induced shift in the distribution of Ras-GTP inactive-active sub-states and recognized potential druggable pockets in the phosphorylated Ras-GTP. Moreover, decreased catalytic competence or signal delivery abilities due to reduced binding affinities and/or distorted catalytic conformations of GAP, SOS and Raf were observed. In addition, the allosteric pathway from Ras/Raf interface to the distal Raf L4 loop was compromised by Ras phosphorylation. These results reveal the mechanisms by which phosphorylation influences the intrinsic or GAP/SOS catalyzed transformations between GTP- and GDP-bound states of Ras and its signal transduction to Raf. Our findings project Ras phosphorylation as a target for cancer drug discovery.

  View details for DOI 10.1016/j.csbj.2021.01.044

  View details for Web of Science ID 000684854500014

  View details for PubMedID 33680360

  View details for PubMedCentralID PMC7902900