Zeyu Zhang

	MedDet: Generative Adversarial Distillation for Efficient Cervical Disc Herniation Detection Zeyu Zhang^∗, Nengmin Yi^∗, Shengbo Tan^∗, Ying Cai^✉, Yi Yang, Lei Xu, Qingtai Li, Zhang Yi, Daji Ergu, Yang Zhao *BIBM 2024* *Oral* Cervical disc herniation (CDH) is a common disorder needing expert analysis. Current automated detection methods face challenges: high computational demands and MRI noise. We propose MedDet for efficient detection, leveraging knowledge distillation, generative adversarial training, and nmODE². Our model improves mAP by 5%, reduces parameters by 67.8%, and speeds inference fivefold.
	Motion Avatar: Generate Human and Animal Avatars with Arbitrary Motion Zeyu Zhang^∗, Yiran Wang^∗, Biao Wu^∗, Shuo Chen, Zhiyuan Zhang, Shiya Huang, Wenbo Zhang, Meng Fang, Ling Chen, Yang Zhao^✉ *BMVC 2024* Our paper introduces a novel agent-based approach called Motion Avatar for generating customizable human and animal 3D avatars with motions via text queries, coordinated by an LLM planner, and supported by the new Zoo-300K animal motion dataset.
	JointViT: Modeling Oxygen Saturation Levels with Joint Supervision on Long-Tailed OCTA Zeyu Zhang, Xuyin Qi, Mingxi Chen, Guangxi Li, Ryan Pham, Ayub Qassim, Ella Berry, Zhibin Liao, Owen Siggs, Robert Mclaughlin, Jamie Craig, Minh-Son To *MIUA 2024* *Oral* Our paper introduces JointViT, a Vision Transformer model with a novel joint loss function and balancing augmentation technique that significantly improves the accuracy of diagnosing sleep-related disorders using OCTA, achieving up to a 12.28% accuracy improvement.
	SegReg: Segmenting OARs by Registering MR Images and CT Annotations Zeyu Zhang, Xuyin Qi, Bowen Zhang, Biao Wu, Hien Le, Bora Jeong, Zhibin Liao, Yunxiang Liu, Johan Verjans, Minh-Son To, Richard Hartley^✉ *ISBI 2024* To improve the accuracy and efficiency of organ at risk (OAR) segmentation in radiotherapy, we propose SegReg, a method that combines CT and MRI using Elastic Symmetric Normalization, outperforming traditional CT-only methods by 16.78% in mDSC and 18.77% in mIoU.
	Thin-Thick Adapter: Segmenting Thin Scans Using Thick Annotations Zeyu Zhang, Bowen Zhang, Abhiram Hiwase, Feng Chen, Akide Liu, Christen Barras, Biao Wu, Adam Wells, Daniel Ellis, Benjamin Reddi, Andrew Burgan, Minh-Son To, Ian Reid, Richard Hartley, Yutong Xie^✉ *Preprint* Medical imaging segmentation is critical for medical analysis, predominantly using thicker CT slices due to the scarcity of annotated thin slices, so we propose segmenting thin scans with thicker slice annotations, introduce the CQ500-Thin dataset, and present the Thin-Thick Adapter to bridge domain gaps, significantly improving segmentation performance.
	DiabetesNet: A Deep Learning Approach to Diabetes Diagnosis Zeyu Zhang, Khandaker Asif Ahmed, Md Rakibul Hasan, Tom Gedeon, Md Zakir Hossain *ACIIDS 2024* We propose a non-invasive diabetes diagnosis method using a Back Propagation Neural Network with batch normalization, addressing class imbalance and improving performance over traditional methods, achieving high accuracy on multiple datasets.

	ProjectedEx: Enhancing Generation in Explainable AI for Prostate Cancer Xuyin Qi^∗, Zeyu Zhang^∗✝, Aaron Berliano Handoko^∗, Huazhan Zheng, Mingxi Chen, Ta Duc Huy, Vu Minh Hieu Phan, Lei Zhang, Linqi Cheng, Shiyu Jiang, Zhiwei Zhang, Zhibin Liao, Yang Zhao^✉, Minh-Son To *Preprint* Prostate cancer diagnosis benefits from MRI and AI advancements. We propose ProjectedEx, a generative framework with multiscale feedback, enhancing interpretability and lesion classification while addressing challenges in medical imaging complexity.

	CT Heterogeneity and Dose Distribution Patterns in Block and Ring Regions Improved the Prediction of Radiation Pneumonitis Yuyu Liu, Wang Li, Fangfang Yang, Yali Wang, Zeyu Zhang, Biao Wu, Gao Yanping, Han Bai, Wenbing Lv *In Submission (EMBC 2025)* This study analyzed 251 lung cancer patients to examine CT heterogeneity and dose distribution's impact on radiation pneumonitis (RP). Using radiomics features and dose-volume histogram parameters across block and ring regions, seven machine learning models were applied. Results showed multi-modal features outperforming DVH, with ring regions (40–50 Gy) achieving the highest AUC (0.977). This supports personalized treatment planning.
	A GNN-based Fraud Detection Approach against Heterophily Inconsistencies Wenxin Zhang, Jingxing Zhong, Zeyu Zhang, Lingfei Ren, Cuicui Luo *In Submission (IJCAI 2025)* Graph-based fraud detection is vital yet challenged by heterophilic inconsistencies, often overlooking semantic nuances. We propose HIGNN, leveraging distinct semantic patterns in heterophilic connections to enhance fraud detection robustness, validated on real-world datasets.
	Bias and Toxicity in Large Language Model Role-Playing Jinman Zhao, Zifan Qian, Linbo Cao, Yining Wang, Yitian Ding, Yulan Hu, Zeyu Zhang *In Submission (ARR)* Role-play in LLMs enhances contextual responses and reasoning across benchmarks but introduces risks, as adopting diverse roles increases susceptibility to biased or harmful outputs on sensitive evaluations.
	GAMED-Snake: Gradient-aware Adaptive Momentum Evolution Deep Snake Model for Multi-organ Segmentation Ruicheng Zhang, Haowei Guo, Zeyu Zhang, Puxin Yan, Shen Zhao *In Submission (ICME 2025)* The Gradient-aware Adaptive Momentum Evolution Deep Snake (GAMED-Snake) model advances multi-organ segmentation, achieving a 2% mDice improvement via novel gradient-based learning, adaptive momentum evolution, and dynamic boundary alignment innovations.
	FDG-Diff: Frequency-Domain-Guided Diffusion Framework for Compressed Hazy Image Restoration Ruicheng Zhang, Kanghui Tian, Zeyu Zhang, Qixiang Liu, Zhi Jin *In Submission (ICME 2025)* This study introduces FDG-Diff, a frequency-domain-guided framework addressing joint haze degradation and JPEG compression. Key innovations include HFCM for detail restoration and DADTP for adaptive region-specific enhancement, achieving superior dehazing performance.
	Class-Centric Semi-supervised Vision Transformers Hongyang He, Haochen You, Zeyu Zhang, Hongyang Xie, Boyang Fu, Guodong Shen, Victor Sanchez *In Submission (ICME 2025)* CLASST, a framework for Semi-Supervised Vision Transformers, addresses class imbalance by dynamically enhancing minority class representation using learnable centers and adaptive contrastive loss, achieving state-of-the-art results on imbalanced datasets.
	AdvMark: Robust Image Watermarking via Two Stage Adversarial Enhancement Jiahui Chen, Zehang Deng, Zeyu Zhang, Chaoyang Li, Lianchen Jia, Lifeng Sun *In Submission (CVPR 2025)* AdvMark introduces a two-stage fine-tuning strategy for robust watermarking, achieving up to 46% accuracy improvement against advanced attacks while ensuring superior image quality through constrained loss and quality-aware early stopping.
	DiffuMural: Diffusion Model for Dunhuang Murals Restoration based on Multi-scale Convergence and Cooperative Diffusion Puyu Han, Jiaju Kang, Yuhang Pan, Erting Pan, Qunchao Jin, Juntao Jiang, Zeyu Zhang, Zhichen Liu, Luqi Gong *In Submission (CVPR 2025)* DiffuMural combines multi-scale convergence, collaborative diffusion, and cyclic loss, excelling in ancient mural restoration with unmatched detail, style coherence, and cultural preservation, surpassing SOTA methods across comprehensive metrics.
	OCRT: Boosting Foundation Models in the Open World with Object-Concept-Relation Triad Luyao Tang, Chaoqi Chen, Yuxuan Yuan, Zeyu Zhang, Yue Huang, Kun Zhang *In Submission (CVPR 2025)* Foundation models struggle with distribution shifts and weak supervision. We propose OCRT, a framework extracting high-level concepts and relations, enhancing SAM and CLIP generalizability in diverse tasks.
	EPDD-YOLO: An Efficient Benchmark for Pavement Damage Detection Based on Mamba-YOLO Shipeng Luo, Yuxin Zhang, Zeyu Zhang, Binhua Guo, Junbo Jacob Lian, Hui Jiang, Shun Zou, Wei Wang *In Submission (Measurement)* EPDD-YOLO enhances pavement damage detection with advanced augmentation, architectural improvements, and the EPDD dataset, achieving 0.873 precision and real-time inference at 198 FPS on challenging benchmarks.
	SegKAN: High-Resolution Medical Image Segmentation with Long-Distance Dependencies Shengbo Tan, Rundong Xue, Shipeng Luo, Zeyu Zhang^✝, Xinran Wang, Lei Zhang, Daji Ergu, Zhang Yi, Yang Zhao, Ying Cai^✉ *Preprint* Hepatic vessel segmentation faces noise and fragmentation challenges. We propose SegKAN, enhancing image embedding and spatial-temporal relationships in Vision Transformers, achieving a 1.78% Dice score improvement on a hepatic vessel dataset.
	Hazards in Daily Life? Enabling Robots to Proactively Detect and Resolve Anomalies Zirui Song, Guangxian Ouyang, Meng Fang, Hongbin Na, Zijing Shi, Zhenhao Chen, Yujie Fu, Zeyu Zhang, Shiyu Jiang, Miao Fang, Ling Chen, Xiuying Chen^✉ *Preprint* Household robots struggle to detect hazards. We propose anomaly scenario generation using multi-agent brainstorming and 3D simulations, enhancing robotic skills in hazard detection, hygiene management, and child safety through diverse environments.
	Medical AI for Early Detection of Lung Cancer: A Survey Guohui Cai, Ying Cai^✉, Zeyu Zhang^✝, Yuanzhouhan Cao, Lin Wu, Daji Ergu, Zhinbin Liao, Yang Zhao *Preprint* Deep learning has revolutionized pulmonary nodule analysis, surpassing traditional methods in detection, segmentation, and classification. This review highlights recent advancements, including CNNs, RNNs, GANs, and ensemble models, improving lung cancer diagnosis.
	MSDet: Receptive Field Enhanced Multiscale Detection for Tiny Pulmonary Nodule Guohui Cai, Ying Cai^✉, Zeyu Zhang^✝, Daji Ergu, Yuanzhouhan Cao, Binbin Hu, Zhinbin Liao, Yang Zhao *Preprint* Pulmonary nodules are vital for early lung cancer diagnosis but challenging to detect. Our proposed MSDet network, with ERD, PCAM, and TODB strategies, achieves state-of-the-art results on LUNA16, improving mAP by 8.8%.
	ESA: Annotation-Efficient Active Learning for Semantic Segmentation Jinchao Ge, Zeyu Zhang, Minh Hieu Phan, Bowen Zhang, Akide Liu, Yang Zhao^✉ *Preprint* Active learning improves annotation efficiency by selecting the most informative samples for labeling. We propose Entity-Superpixel Annotation (ESA), an efficient strategy using a mask proposal network and superpixel grouping. Our method reduces click cost by 98% and boosts performance by 1.71%, outperforming pixel-based methods with only 40 clicks per image.
	SegStitch: Multidimensional Transformer for Robust and Efficient Medical Imaging Segmentation Shengbo Tan, Zeyu Zhang^✝, Ying Cai^✉, Daji Ergu, Lin Wu, Binbin Hu, Pengzhang Yu, Yang Zhao *Preprint* Medical imaging segmentation, crucial for lesion analysis, has seen advances with transformers in 3D segmentation. Despite their scalability, transformers struggle with local features and complexity. We propose SegStitch, combining transformers with denoising ODE blocks, improving mDSC by up to 11.48% and reducing parameters by 36.7%, promising real-world clinical adaptation.
	Sine Activated Low-Rank Matrices for Parameter Efficient Learning Yiping Ji^∗, Hemanth Saratchandran^∗, Cameron Gordon, Zeyu Zhang, Simon Lucey^✉ *Preprint* We propose a novel theoretical framework integrating a sinusoidal function into low-rank decomposition, enhancing parameter efficiency and model accuracy across diverse neural network applications such as Vision Transformers, Large Language Models, Neural Radiance Fields, and 3D shape modeling.
	XLIP: Cross-modal Attention Masked Modelling for Medical Language-Image Pre-Training Biao Wu^∗, Yutong Xie^∗, Zeyu Zhang, Minh Hieu Phan, Qi Chen, Ling Chen, Qi Wu^✉ *Preprint* Vision-and-language pretraining (VLP) in the medical field faces challenges with reconstructing pathological features due to data scarcity and limited use of paired/unpaired data. This paper proposes XLIP, using AttMIM and EntMLM modules, to enhance feature learning from unpaired data, achieving state-of-the-art results in medical classification tasks.
	A Landmark-Based Approach for Instability Prediction in Distal Radius Fractures Yang Zhao, Zhibin Liao, Yunxiang Liu, Koen Oude Nijhuis, Britt Barvelink, Jasper Prijs, Joost Colaris, Mathieu Wijffels, Max Reijman, Zeyu Zhang, Minh-Son To, Ruurd Jaarsma, Job Doornberg, Johan Verjans *ISBI 2024* Distal radius fractures (DRFs) are common and their instability assessment is crucial for treatment decisions, affecting recovery and costs. We propose a deep learning-based landmark detection method using anatomical landmarks from X-rays to measure distances and angles. These features are used in an XGBoost model for DRF instability classification, validated on a large Dutch dataset.
	Can Rotational Thromboelastometry Rapidly Identify Theragnostic Targets in Isolated Traumatic Brain Injury? Abhiram Hiwase, Christopher Ovenden, Lola Kaukas, Mark Finnis, Zeyu Zhang, Stephanie O'Connor, Ngee Foo, Benjamin Reddi, Adam Wells, Daniel Ellis *EMA 2024* This study evaluates the prognostic utility of ROTEM sigma in isolated traumatic brain injury (TBI). ROTEM sigma, a point-of-care assay, demonstrated faster turnaround times and comparable accuracy to standard coagulation tests in predicting head injury-related deaths. The findings suggest ROTEM sigma effectively detects coagulopathy in isolated TBI cases.
	BHSD: A 3D Multi-class Brain Hemorrhage Segmentation Dataset Biao Wu, Yutong Xie, Zeyu Zhang, Jinchao Ge, Kaspar Yaxley, Suzan Bahadir, Qi Wu, Yifan Liu, Minh-Son To^✉ *MLMI 2023* The Brain Hemorrhage Segmentation Dataset (BHSD) is a comprehensive 3D multi-class ICH dataset with pixel-level and slice-level annotations designed to support supervised and semi-supervised ICH segmentation tasks, addressing the lack of existing public datasets for multi-class ICH segmentation.

News

Publications

Research Experience

Education

Honors & Awards

Academic Services

Talks