Maggie Shin-Young Chen

All Publications

• Independent Evaluation of RETFound Foundation Model's Performance on Optic Nerve Analysis Using Fundus Photography. Ophthalmology science Chen, M. S., Ravindranath, R., Chang, R., Zhou, Y., Keane, P. A., Wang, S. Y. 2025; 5 (3): 100720

  Abstract

  This study evaluates RETFound, a retinal image foundation model, as a feature extractor for predicting optic nerve metrics like cup-to-disc ratio (CDR) and retinal nerve fiber layer (RNFL) thickness using an independent clinical dataset.Retrospective observational study.Patients who underwent fundus photography and RNFL OCT at the Byers Eye Institute, Stanford University.Fundus images were paired with RNFL OCT results where study dates were within 6 months of each other. Latent features from full-sized raw fundus images were extracted from RETFound and used as inputs for several linear regression models (Ridge, Lasso, Elastic Net, and ordinary least squares). Baseline models using pretrained VGG16 and Vision Transformers (ViTs) as feature extractors were also developed. All models were trained to perform single-output tasks (predicting CDR or average RNFL thickness) and multioutput tasks (predicting RNFL thickness at quadrants and clock hours). Data were split 80:20 at the patient level for training and validation.Model predictions were evaluated on a test set using the metrics of R 2 , mean absolute error, and root mean square error.Among the 463 unique participants, contributing 776 fundus-OCT data pairs, the mean age was 63 years (±18 years), with 57.24% being female (N = 265). RETFound models demonstrated strong performance on single-output tasks, achieving R 2 values between 0.706 and 0.898 for CDR prediction and between 0.855 and 0.961 for average RNFL thickness prediction. Performance on multioutput tasks was less robust, with a highest R 2 of 0.583 for clock-hour RNFL thickness prediction and an R 2 of 0.811 for quadrant RNFL thickness prediction. RETFound models outperformed VGG16 and ViT models, which achieved maximum R 2 of 0.731 and 0.687 in predicting RNFL thickness and CDR.Machine learning models leveraging the massively pretrained RETFound foundation model could accurately predict CDR and average RNFL thickness from fundus photos on an independent clinical dataset. Although RETFound was not trained or fine-tuned for these optic nerve evaluation tasks, nevertheless, RETFound overcomes small dataset limitations and excels in specialized applications.Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

  View details for DOI 10.1016/j.xops.2025.100720

  View details for PubMedID 40161459

  View details for PubMedCentralID PMC11950761

• Genomic predictors of radiation response: recent progress towards personalized radiotherapy for brain metastases. Cell death discovery Harary, P. M., Rajaram, S., Chen, M. S., Hori, Y. S., Park, D. J., Chang, S. D. 2024; 10 (1): 501

  Abstract

  Radiotherapy remains a key treatment modality for both primary and metastatic brain tumors. Significant technological advances in precision radiotherapy, such as stereotactic radiosurgery and intensity-modulated radiotherapy, have contributed to improved clinical outcomes. Notably, however, molecular genetics is not yet widely used to inform brain radiotherapy treatment. By comparison, genetic testing now plays a significant role in guiding targeted therapies and immunotherapies, particularly for brain metastases (BM) of lung cancer, breast cancer, and melanoma. Given increasing evidence of the importance of tumor genetics to radiation response, this may represent a currently under-utilized means of enhancing treatment outcomes. In addition, recent studies have shown potentially actionable mutations in BM which are not present in the primary tumor. Overall, this suggests that further investigation into the pathways mediating radiation response variability is warranted. Here, we provide an overview of key mechanisms implicated in BM radiation resistance, including intrinsic and acquired resistance and intratumoral heterogeneity. We then discuss advances in tumor sampling methods, such as a collection of cell-free DNA and RNA, as well as progress in genomic analysis. We further consider how these tools may be applied to provide personalized radiotherapy for BM, including patient stratification, detection of radiotoxicity, and use of radiosensitization agents. In addition, we describe recent developments in preclinical models of BM and consider their relevance to investigating radiation response. Given the increase in clinical trials evaluating the combination of radiotherapy and targeted therapies, as well as the rising incidence of BM, it is essential to develop genomically informed approaches to enhance radiation response.

  View details for DOI 10.1038/s41420-024-02270-2

  View details for PubMedID 39695143

  View details for PubMedCentralID PMC11655559

• Fillable Magnetic Microrobots for Drug Delivery to Cardiac Tissues In Vitro. Advanced healthcare materials Chen, M. S., Sun, R., Wang, R., Zuo, Y., Zhou, K., Kim, J., Stevens, M. M. 2024; 13 (22): e2400419

  Abstract

  Many cardiac diseases, such as arrhythmia or cardiogenic shock, cause irregular beating patterns that must be regulated to prevent disease progression toward heart failure. Treatments can include invasive surgery or high systemic drug dosages, which lack precision, localization, and control. Drug delivery systems (DDSs) that can deliver cargo to the cardiac injury site could address these unmet clinical challenges. Here, a microrobotic DDS that can be mobilized to specific sites via magnetic control is presented. This DDS incorporates an internal chamber that can protect drug cargo. Furthermore, the DDS contains a tunable thermosensitive sealing layer that gradually degrades upon exposure to body temperature, enabling prolonged drug release. Once loaded with the small molecule drug norepinephrine, this microrobotic DDS modulated beating frequency in induced pluripotent stem-cell derived cardiomyocytes (iPSC-CMs) in a dose-dependent manner, thus simulating drug delivery to cardiac cells in vitro. The DDS also navigates several maze-like structures seeded with cardiomyocytes to demonstrate precise locomotion under a rotating low-intensity magnetic field and on-site drug delivery. This work demonstrates the utility of a magnetically actuating DDS for precise, localized, and controlled drug delivery which is of interest for a myriad of future opportunities such as in treating cardiac diseases.

  View details for DOI 10.1002/adhm.202400419

  View details for PubMedID 38748937

• Senescence mechanisms and targets in the heart. Cardiovascular research Chen, M. S., Lee, R. T., Garbern, J. C. 2022; 118 (5): 1173-1187

  Abstract

  Cellular senescence is a state of irreversible cell cycle arrest associated with ageing. Senescence of different cardiac cell types can direct the pathophysiology of cardiovascular diseases (CVDs) such as atherosclerosis, myocardial infarction, and cardiac fibrosis. While age-related telomere shortening represents a major cause of replicative senescence, the senescent state can also be induced by oxidative stress, metabolic dysfunction, and epigenetic regulation, among other stressors. It is critical that we understand the molecular pathways that lead to cellular senescence and the consequences of cellular senescence in order to develop new therapeutic approaches to treat CVD. In this review, we discuss molecular mechanisms of cellular senescence, explore how cellular senescence of different cardiac cell types (including cardiomyocytes, cardiac endothelial cells, cardiac fibroblasts, vascular smooth muscle cells, and valve interstitial cells) can lead to CVD, and highlight potential therapeutic approaches that target molecular mechanisms of cellular senescence to prevent or treat CVD.

  View details for DOI 10.1093/cvr/cvab161

  View details for PubMedID 33963378

  View details for PubMedCentralID PMC8953446

• Emerging Approaches to Functionalizing Cell Membrane-Coated Nanoparticles. Biochemistry Ai, X., Wang, S., Duan, Y., Zhang, Q., Chen, M. S., Gao, W., Zhang, L. 2021; 60 (13): 941-955

  Abstract

  There has been significant interest in developing cell membrane-coated nanoparticles due to their unique abilities of biomimicry and biointerfacing. As the technology progresses, it becomes clear that the application of these nanoparticles can be drastically broadened if additional functions beyond those derived from the natural cell membranes can be integrated. Herein, we summarize the most recent advances in the functionalization of cell membrane-coated nanoparticles. In particular, we focus on emerging methods, including (1) lipid insertion, (2) membrane hybridization, (3) metabolic engineering, and (4) genetic modification. These approaches contribute diverse functions in a nondisruptive fashion while preserving the natural function of the cell membranes. They also improve on the multifunctional and multitasking ability of cell membrane-coated nanoparticles, making them more adaptive to the complexity of biological systems. We hope that these approaches will serve as inspiration for more strategies and innovations to advance cell membrane coating technology.

  View details for DOI 10.1021/acs.biochem.0c00343

  View details for PubMedID 32452667

  View details for PubMedCentralID PMC8507422

• Fabrication and characterization of a 3D bioprinted nanoparticle-hydrogel hybrid device for biomimetic detoxification. Nanoscale Chen, M. S., Zhang, Y., Zhang, L. 2017; 9 (38): 14506-14511

  Abstract

  A biomimetic micro/nanodevice is 3D bioprinted using polyethylene glycol (PEG) hydrogel as the supporting platform, along with the red blood cell (RBC) membrane-coated nanoparticles (RBC-NPs) encapsulated in the hydrogel as the detoxification mechanism. RBC-NPs are prepared through a nanoprecipitation and coating method and then mixed into the poly(ethylene glycol) diacrylate (PEGDA) monomer solution for 3D bioprinting through photopolymerization. This resulting detoxification device is engineered with multiple inner channels for the RBC-NPs to nonspecifically soak up the various toxins flowing through the channels. Different shapes (i.e. star or triangle) of the channel are fabricated, each with a larger surface area than the generic circle shape. The device is characterized for microstructure, nanoparticle encapsulation and function, and its detoxification ability. Overall, the strategy of incorporating functional nanoparticles into a biocompatible hydrogel as the supporting platform may enable localized, patient specific controlled therapeutics for detoxification, drug delivery, and other precision medicine application.

  View details for DOI 10.1039/c7nr05322c

  View details for PubMedID 28930358

• A Bioadhesive Nanoparticle-Hydrogel Hybrid System for Localized Antimicrobial Drug Delivery. ACS applied materials & interfaces Zhang, Y., Zhang, J., Chen, M., Gong, H., Thamphiwatana, S., Eckmann, L., Gao, W., Zhang, L. 2016; 8 (28): 18367-74

  Abstract

  Effective antibacterial treatment at the infection site associated with high shear forces remains challenging, owing largely to the lack of durably adhesive and safe delivery platforms that can enable localized antibiotic accumulation against bacterial colonization. Inspired by delivery systems mimicking marine mussels for adhesion, herein, we developed a bioadhesive nanoparticle-hydrogel hybrid (NP-gel) to enhance localized antimicrobial drug delivery. Antibiotics were loaded into polymeric nanoparticles and then embedded into a 3D hydrogel network that confers adhesion to biological surfaces. The combination of two distinct delivery platforms, namely, nanoparticles and hydrogel, allows the hydrogel network properties to be independently tailored for adhesion while maintaining controlled and prolonged antibiotic release profile from the nanoparticles. The bioadhesive NP-gel developed here showed superior adhesion and antibiotic retention under high shear stress on a bacterial film, a mammalian cell monolayer, and mouse skin tissue. Under a flow environment, the NP-gel inhibited the formation of an Escherichia coli bacterial film. When applied on mouse skin tissue for 7 consecutive days, the NP-gel did not generate any observable skin reaction or toxicity, implying its potential as a safe and effective local delivery platform against microbial infections.

  View details for DOI 10.1021/acsami.6b04858

  View details for PubMedID 27352845

  View details for PubMedCentralID PMC4983189