About AI4Rad Lab
Based in the Department of Radiology at Shanghai Sixth People's Hospital, Shanghai Jiao Tong University School of Medicine, our group unites clinicians, engineering researchers, postdoctoral fellows, graduate students, and SJTU-SM undergraduate RBL / innovation trainees around medical imaging artificial intelligence.
Mission & Research Goals
Methods Grounded in Clinical Practice
We study how AI extracts diagnostically meaningful representations from CT, MRI, and multimodal clinical records to support disease detection, risk stratification, emergency triage, and structured reporting — evaluating reliability through external cohorts and radiologist reader studies.
Quantitative Phenotypes & Mechanisms
Beyond diagnostic classification, we investigate how high-resolution reconstruction, 3D anatomical modeling, vascular hemodynamics, and musculoskeletal biomechanics link imaging phenotypes with underlying physiological and pathological processes.
Research Platform & Funding Support
Our lab operates dedicated in-house high-performance GPU computing and petabyte-scale clinical imaging storage, backed by sustained national research funding to support foundation-model training, multicenter cohort curation, and student research.
In-House High-Performance Compute
Equipped with a dedicated 100-GPU-scale high-performance computing cluster (>100 kW power capacity), enabling queue-free self-supervised pretraining on 100,000+ 3D CT/MRI scans, multimodal LLM fine-tuning, and large-scale biomechanical simulation.
Petabyte-Scale Imaging Storage
Supported by dedicated petabyte-scale storage arrays for secure governance and high-throughput training across multimodal clinical cohorts, decade-long longitudinal follow-up studies, and paired clinical CT / photon-counting micro-CT datasets.
National Projects & Sustained Funding
Over the past five years, team investigators have led more than 10 national-level research grants with total funding exceeding ¥10 million RMB — providing strong support for graduate training, SJTU-SM undergraduate RBL / innovation projects, and multicenter studies.
Clinical Imaging Directions
Our four methodological pillars — vision foundation models, multimodal large language models, image generation & reconstruction, and spatial intelligence & physical simulation — are organized around seven clinical radiology domains at Shanghai Sixth People's Hospital.
- Pan-Vascular & Hemodynamic Imaging Cerebrovascular, coronary, carotid, aortic, and peripheral vascular anatomy, perfusion, thrombus characterization, and hemodynamic risk events.
- Musculoskeletal & Bone Quantitative Imaging Spine, joints, tendons, fractures, trabecular micro-architecture, and quantitative bone metabolism across clinical CT, MRI, and micro-CT.
- Oncologic Imaging & Whole-Body Metastasis Imaging-based tumor detection, staging, whole-body skeletal metastasis assessment, treatment response, and radiologic–pathologic correlation.
- Emergency & Acute Care Imaging Time-critical head, chest, and non-contrast acute abdominal triage, intracranial trauma risk assessment, and emergency decision support.
- Longitudinal Chronic Disease Cohorts Mining multi-timepoint hospital imaging cohorts to model disease progression, biomarker trajectories, and long-term clinical outcomes.
- Radiology Education & Reporting Feedback Criteria-guided structured reporting, diagnostic impression evaluation, quality assurance, and human–AI collaborative reader workflows.
- Emerging Imaging Modalities & Translation Clinical evaluation and computational methods for photon-counting CT, dual-energy spectral CT, accelerated MRI, and 4D flow imaging.
How We Work
Clinician–Engineer Pairing
Radiologists and AI researchers collaborate from problem formulation and cohort curation through task-chain design and error analysis.
Multicenter & Reader Studies
Models are tested across independent clinical centers and structured multi-reader studies rather than single-center retrospective splits alone.
Benchmarks & Reproducibility
We release curated benchmarks, paired datasets, and open-source reference implementations so the community can verify and build on our work.
Explore our Team, browse peer-reviewed work under Publications, or reach out via Contact.