Biology subjects

Leng, T.

Publications and source records attributed to Leng, T..

7 recordsLinked to original sources

bioRxiv2025

Compact Vision--Language Models Enable Efficient and Interpretable Automated OCT Analysis Through Layer Specific Multimodal Learning

Translating the intricate anatomical signatures of retinal disease from OCT B-scans into clear, accurate clinical narratives demands AI models that seamlessly fuse visual features with domain expertise. We curated a multimodal dataset of 40,000 OCT B-scans from public repositories and private clinical cohorts, each paired with expert validated summaries spanning six conditions: diabetic macular edema, diabetic retinopathy, geographic atrophy, drusen, choroidal neovascularization, and healthy retina. We introduce LO-VLM, a compact (247M parameter) vision-language model (VLM) that infuses anatomical guidance into both encoder and decoder for free form summary generation and multiclass disease classification. Benchmarking against state-of-the-art RetinaVLM, LLaVA-Med, and a ViT vision only model demonstrates superior performance. In a blinded evaluation by three board certified retina specialists scored the generated summaries, LO-VLM narratives achieved mean = 8.5 (standard deviation = 1.15) out of 10, compared to mean = 5.5 (standard deviation = 1.13) for RetinaVLM (p < 0.0001). In quantitative evaluations, LO-VLM achieved an SBERT similarity of 0.803 and a BERTScore F1 of 0.715, representing improvements of 8.2% and 28.8% over specialized VLM baselines. For disease classification, LO-VLM reached 96% accuracy (F1 = 96%), outperforming ViT by 13% and exceeding medical VLM benchmarks by over 62%. By reconciling interpretability with computational efficiency, LO-VLM establishes a new paradigm for efficient AI models in OCT interpretation.

bioengineering↗

bioRxiv2024

EYE-Llama, an in-domain large language model for ophthalmology

1BackgroundTraining Large Language Models (LLMs) with in-domain data can significantly enhance their performance, leading to more accurate and reliable question-answering (Q&A) systems essential for supporting clinical decision-making and educating patients. MethodsThis study introduces ophthalmic LLMs trained on in-domain, well-curated datasets. We present an open-source substantial ophthalmic language dataset for model training. Our models (EYE-Llama), were pre-trained on an ophthalmology-specific dataset, including paper abstracts, textbooks, and Wikipedia articles. Subsequently, the models underwent fine-tuning using a diverse range of QA pairs. Our models were compared to baseline Llama 2, ChatDoctor, Meditron, Llama 3, and ChatGPT (GPT3{middle dot}5) models, using four distinct test sets, and evaluated quantitatively (Accuracy, F1 score, BERTScore, BARTScore and BLEU score) and qualitatively by two ophthalmologists. FindingsUpon evaluating the models using the synthetic dialogue test set with three different metrics (BERTScore, BARTScore, and BLEU score), our models demonstrated superior performance. Specifically, when evaluated using BERTScore, our models surpassed Llama 2, Llama 3, Meditron, and ChatDoctor in terms of F1 score, and performed on par with ChatGPT, which was trained with 175 billion parameters (EYE-Llama: 0.57, Llama 2: 0.56, Llama 3: 0.55, Meditron: 0.50, ChatDoctor: 0.56, and ChatGPT: 0.57). Additionally, the EYE-Llama model outperformed the above models when evaluated using BARTScore and BLEU scores. When tested on the MedMCQA test set, the fine-tuned models exhibited higher accuracy compared to Llama 2, Meditron, and ChatDoctor models (EYE-Llama: 0.39, Llama 2: 0.33, ChatDoctor: 0.29, Meditron: 0.22). However, ChatGPT, and Llama 3 models outperformed EYE-Llama, achieving accuracies of 0.55, 0.78, and 0.90, respectively. On the PubmedQA test set, our model showed improved accuracy over all other models (EYE-Llama: 0.96, Llama 2: 0.90, Llama 3: 0.92, Meditron: 0.76, ChatGPT: 0.93, ChatDoctor: 0.92). InterpretationThe study shows that pre-training and fine-tuning LLMs like EYE-Llama enhances their performance in specific medical domains. Our EYE-Llama models surpass baseline Llama 2 in all evaluations, highlighting the effectiveness of specialized LLMs in medical QA systems. FundingFunded by NEI R15EY035804 (MNA), R21EY035271 (MNA), and UNC Charlotte Faculty Research Grant (MNA)

bioengineering↗

bioRxiv2023

Federated learning for diagnosis of age-related macular degeneration

This paper presents a federated learning (FL) approach to train deep learning models for classifying age-related macular degeneration (AMD) using optical coherence tomography image data. We employ the use of residual networks and vision transformer encoders for the normal vs AMD binary classification, integrating four unique domain adaptation techniques to address domain shift issues caused by heterogeneous data distribution in different institutions. Experimental results indicate that FL strategies can achieve competitive performance similar to centralized models even though each local model has access to a portion of the training data. Notably, Adaptive Personalization FL strategy stood out in our FL evaluations, consistently delivering high performance across all tests due to its additional local model. Furthermore, the study provides valuable insights into the efficacy of simpler architectures in image classification tasks, particularly in scenarios where data privacy and decentralization are critical using both encoders. It suggests future exploration into deeper models and other FL strategies for a more nuanced understanding of these models performance.

bioengineering↗

bioRxiv2023

CHPF boosts proliferation and invasion of clear cell renal cell carcinoma

BackgroundClear cell renal cell carcinoma (ccRCC) is a malignant tumor most commonly seen in the urinary system, featuring quick progression, invasive behavior, frequent recrudesce, and bad prognosis, whereas Chondroitin polymerizing factor (CHPF) is an essential glycosyl transferase of biosynthesis involved chondroitin sulphate. But the relationship between the two has not been fully understood so far. The present research will probe into the relationship between CHPF and ccRCC. MethodsExplore the CHPF expression level in renal ccRCC tissues through bioinformatics analysis of The Cancer Genome Atlas (TCGA) and the real-time quantitative polymerase chain reaction detection system (qPCR); acquire relevant clinical data from the TCGA data-base and use Kaplan-Meier survival analysis to verify the relevance of CHPF expression to the clinical prognosis of ccRCC patients; then, effectively silence CHPF in ccRCC 786-O cells by a lentivirus-mediated approach; next, observe the effects of CHPF on tumor cell proliferation, cell cycle progression, and apoptosis. ResultsIn ccRCC tissues, CHPF expression has been significantly upregulated; the higher expression, the shorter survival of ccRCC patients. CHPF downregulation in the 786-O cell can effectively hold back the proliferation, apoptosis and cycle of cancer cells. ConclusionBased on the above results, we arrive at a conclusion that CHPF expression contributes markedly to the development of human ccRCC cells. Therefore, CHPF may act as a potential prognostic marker of ccRCC and provide a new target for ccRCC treatment. Simple SummaryChondroitin polymerizing factor (CHPF) is an important glycosyltransferase involved in the biosynthesis of chondroitin sulfate, which is thought to have some pro-carcinogenic effects. However, the role and mechanism of CHPF in clear cell renal cell carcinoma (ccRCC) have not been reported. The aim of this study was to investigate the relationship between CHPF and ccRCC. In this study, we found that CHPF was highly expressed in ccRCC and knocking out CHPF can greatly inhibit the proliferation and cell cycle progression of ccRCC and accelerate its apoptosis, suggesting that CHPF can predict the prognosis of ccRCC and a potential target for treatment.

genetics↗

bioRxiv2018

TCR and inflammatory signals tune human MAIT cells to exert specific tissue repair and effector functions

MAIT cells are an abundant T cell population that are highly enriched in epithelial tissues such as the liver. They are known to be activated both through TCR-dependent and -independent mechanisms. However, the specific functional responses of MAIT cells to these signals remain elusive. We examined the impact of combinations of TCR-dependent and -independent signals in blood and tissue-derived human MAIT cells. TCR-dependent and -independent signals drive MAIT cells to exert overlapping and unique effector functions. Importantly, this includes several novel functions that are relevant not only to host defence but also tissue homeostasis. While TCR triggering is insufficient to drive sustained activation, gene expression signatures of TCR-triggered MAIT cells also showed specific enrichment of tissue-repair functions. Activation of certain effector and antimicrobial functions may be triggered further with specific innate cytokine signals, such as the TNF superfamily member TL1A. In addition, gut-resident cells exposed to commensal-derived ligands showed evidence of persistent activation in vivo. Taken together, this study demonstrates the pivotal role of MAIT cells in controlling the balance between healthy and pathological processes of tissue inflammation and repair.

immunology↗

bioRxiv2018

Type-2 CD8+ T lymphocytes responsive to PGD2 and LTE4 in severe eosinophilic asthma

The functions and in vivo roles of type-2 CD8+ T cells in humans have not been well defined and this cell type has been largely overlooked in models of disease. We investigated this in the context of severe asthma with persistent airway eosinophilia - a phenotype associated with high exacerbation risk and responsiveness to type-2 cytokine-targeted therapies. In two independent cohorts we show that, in contrast to Th2 cells, type-2 cytokine-secreting CD8+CRTH2+ (Tc2) cells are enriched in blood and airways in severe eosinophilic asthma. Concentrations of prostaglandin D2 (PGD2) and cysteinyl leukotriene E4 (LTE4) are also increased in the airways of the same group of patients. In vitro PGD2 and LTE4 function synergistically to trigger Tc2 cell recruitment and activation in a TCR-independent manner. These lipids regulate diverse genes in Tc2 cells inducing type-2 cytokines and many other pro-inflammatory cytokines and chemokines which could contribute to eosinophilia. These findings are consistent with an important innate-like role for human Tc2 cells in severe eosinophilic asthma and suggest a potential target for therapeutic intervention in this and other diseases.

immunology↗

bioRxiv2017

Diverse Streptococcus pneumoniae strains drive a MAIT cell response through MR1-dependent and cytokine-driven pathways

Mucosal Associated Invariant T (MAIT) cells represent an innate T cell population of emerging significance. These abundant cells can recognize ligands generated by microbes utilizing the riboflavin synthesis pathway, presented via the major histocompatibility complex (MHC) class I-related molecule MR1 and binding of specific T cell receptors (TCR). They also possess an innate functional programme allowing microbial sensing in a cytokine-dependent, TCR-independent manner. Streptococcus pneumoniae is a major human pathogen that is also associated with commensal carriage, thus host control at the mucosal interface is critical. The recognition of S. pneumoniae strains by MAIT cells has not been defined, nor have the genomics and transcriptomics of the riboflavin operon (Rib genes). We examined the expression of Rib genes in S. pneumoniae at rest and in response to metabolic stress and linked this to MAIT cell activation in vitro. We observed robust recognition of S. pneumoniae strains at rest and following stress, using both TCR-dependent and TCR-independent pathways. The pathway used was highly dependent on the antigen-presenting cell, but was maintained across a wide range of clinically-relevant strains. The riboflavin operon was highly conserved across a range of 571 S. pneumoniae from 39 countries dating back to 1916, and different versions of the riboflavin operon were also identified in related Streptococcus species. These data indicate an important functional relationship between MAIT cells and S. pneumoniae,which may be tuned by local factors, including the metabolic state of the organism and the antigen-presenting cell that it encounters.\n\nAuthor SummaryStreptococcus pneumoniae is the leading cause of bacterial pneumonia, causes invasive diseases such as meningitis and bacteraemia, and is associated with significant morbidity and mortality, particularly in children and the elderly. Here, we demonstrate that a novel T cell population called Mucosal-associated invariant T (MAIT) cells is able to respond to a diverse range of S. pneumoniae strains. We found that this response was dependent on the T cell receptor (which recognises metabolites of the bacterial riboflavin biosynthesis pathway), cytokines, and the type of antigen-presenting cell. A population genomics approach was also used to assess the prevalence and diversity of the genes encoding the riboflavin biosynthesis pathway among a large and diverse collection of S. pneumoniae. These genes were highly conserved across a range of 571 S. pneumoniae from 39 countries dating back to 1916, and was also present in other related Streptococcus species. Given the low levels of MAIT cells in neonates and MAIT cell decline in the elderly, both of whom are at the highest risk of invasive pneumococcal disease, further understanding of the functional role of MAIT cells in host defense against this major pathogen may allow novel therapeutics or vaccines to be designed.

immunology↗