Hongxi Yang
I am a final-year PhD candidate in the Department of Data Science and AI, Faculty of IT, Monash University, working in the AIM for Health Lab under the supervision of A/Prof. Zongyuan Ge, Dr. Yasmeen George, and Dr. Deval Mehta. My research focuses on multi-modal deep learning for acute ischemic stroke management, including CT perfusion-based lesion prediction, semantic-guided image synthesis, and interpretable clinical decision support. I also serve as a reviewer for MICCAI.
I am a final-year PhD candidate in the AIM for Health Lab (led by A/Prof. Zongyuan Ge) at Monash University's Faculty of IT. My research sits at the intersection of computer vision, generative AI, and clinical neuroscience.
I play a key technical role in the CRC-P project "AI-powered clinical decision assistance for acute stroke" — a national multi-institution collaboration with Apollo Medical Imaging Technology, Ingham Institute, and Queensland Health. My work spans multi-centre data curation across imaging modalities (CTP, CTA, NCCT) and clinical data, developing end-to-end models for lesion segmentation, large vessel occlusion detection, and treatment outcome prediction, to be validated across 17 hospital sites.
Prior to my PhD, I completed my MSc and BSc at East China Normal University, where I worked on radiomics and deep learning for glioma diagnosis. I also interned at Siemens Healthineers, building cerebral vasculature segmentation and defect inspection systems.
Research Interests
Technical Skills
AI-Powered Stroke Decision Support
Key technical contributor to a national CRC-P project with industry and clinical partners. Developed end-to-end pipelines for ischemic lesion segmentation, large vessel occlusion detection, and treatment outcome prediction from multi-modal imaging (CTP, CTA, NCCT) and clinical data — to be validated across 17 hospital sites.
Semantic-Guided 3D CT Synthesis
Developed a latent diffusion model for synthesizing 3D CT images with controllable stroke lesions. Designed a Spatial-Semantic Feature Interaction Module integrating DINOv2 features with adapters, improving downstream segmentation by up to 22% in Dice.
Logic-Guided Diagnostic Reasoning
Built an interpretable framework for medical image classification using logic Concept Bottleneck Models to construct Chain-of-Thought reasoning chains, fine-tuning LLMs (Qwen3-VL, LLaVA, LLaMA) via LoRA for clinically grounded diagnosis.
Glioma Radiomics & Deep Learning
Developed multi-task radiomics and deep learning models for predicting epilepsy and IDH mutation status in glioma patients from preoperative MRI, with shared feature representations and comprehensive preprocessing pipelines.
PhD in Data Science & Artificial Intelligence
DTI Intern
MSc in Radio Physics
BSc in Physics
I'm always open to research collaborations, especially in medical AI, multi-modal learning, and stroke neuroimaging. Feel free to reach out if you'd like to discuss ideas or potential projects.
Say Hello →