🌐 中文  ·  English

AI4Rad Lab · Medical Imaging AI

Medical Imaging Artificial Intelligence Laboratory
Department of Radiology, Shanghai Sixth People's Hospital · Shanghai Jiao Tong University School of Medicine

🌐 ai4rad-ssph.github.io

---

The official website of AI4Rad Lab, a medical imaging AI research group at the Department of Radiology, Shanghai Sixth People's Hospital (SJTU School of Medicine).

We build AI that works in real clinical radiology — from vessel segmentation and perfusion analysis to large-scale disease diagnosis and automated reporting. Our core clinical directions are pan-vascular imaging (cerebrovascular, cardiac & coronary, and peripheral vasculature) and musculoskeletal imaging, alongside broader radiology applications, all driving the clinical translation of medical AI.

🔬 Research directions

Our research is built on four core AI capabilities, applied across the lab's seven clinical imaging directions.

• Vision foundation models — visual representation and structured understanding of medical images: diagnosis, segmentation, detection, and quantification.
• Multimodal large language models — integrating images, reports, clinical history, and knowledge for diagnostic reasoning and structured reporting.
• Medical image generation — clinically useful synthesis, restoration, denoising, and enhancement, including cross-modal generation.
• Spatial intelligence & physical simulation — 3D modeling, registration, blood-flow and musculoskeletal mechanics, and interpretable quantitative analysis.

⭐ Selected projects

Project	Venue	Links
Imaging-Grounded Stroke MLLM — multi-agent multimodal LLM framework for imaging-grounded acute ischemic stroke treatment recommendation	npj Digital Medicine, 2026	Paper
HR-LLM-Stroke — multi-agent LLM framework for emergency stroke treatment recommendation	J Med Internet Res, 2026	Paper · Code · Demo
BrainMIND — multicenter benchmark & reader study evaluating 10 LLMs on brain MRI diagnostic impression generation	npj Digital Medicine, 2026	Paper · Code
LumbarSR — paired clinical CT & micro-CT dataset for lumbar vertebra super-resolution	Scientific Data, 2026	Paper · Challenge
Cerebrovascular & Stroke AI — aneurysm detection, ASPECTS scoring, thrombus characterization	Radiology / European Radiology	—

See the full list at Publications.

🏗️ Repository structure

.
├── index.html / index_ch.html        # Homepage (EN / 中文)
├── about.html / about_ch.html         # About — mission & vision
├── team.html  / team_ch.html          # Team page, rendered from data/team.json
├── publications.html / _ch.html       # Filterable publications list (lab papers)
├── contact.html / _ch.html            # Contact page
├── members/
│   └── index.html                     # Shared member-page template (?id=<id>&lang=cn)
├── css/style.css                      # Styles, dark mode, responsive 2K/4K
├── js/
│   ├── main.js                        # Theme toggle, scroll reveal, team renderer
│   ├── pub.js                         # Publication card renderer + filters + author links
│   └── home.js                        # Shared bilingual news and expandable archive
├── data/
│   ├── news.json                      # Homepage announcements (EN / 中文)
│   ├── team.json                      # Team roster (PI-maintained, 39 members)
│   ├── members/<id>.json              # One file per member — owned by that member
│   ├── publications/<paper-id>.json   # One file per paper + auto-generated manifest.json
│   └── members/_template.json         # Copy to start your own page
├── tools/build-pubs.py                # Rebuild publications manifest (validates fields)
├── images/
│   ├── brand/                         # Lab logo + SSPH/SJTU affiliation logos
│   ├── members/                       # Member photos
│   └── *.png/jpg                      # Paper cover figures
├── MEMBER_GUIDE.md                    # How members update their own info & page
└── DESIGN_SYSTEM.md                   # Visual identity spec (colors, fonts, logos)

⚙️ How it works

• Data-driven. Pages fetch() JSON and render with vanilla JS — no build step, no framework. Everything is static and served by GitHub Pages (with .nojekyll).
• News reads data/news.json for both languages. Five current announcements are visible; earlier news is expandable.
• Publication status distinguishes accepted work from online and published articles. Known publication dates order papers within each year; filters also support status and Chinese author names.
• Frontend checks run with node tools/test-site.js and in PR CI. Shared CSS/JS URLs carry a dated version; bump it across pages when changing these assets.
• Team page reads data/team.json; a member's card links to a personal page once their entry has "page": true.
• Member pages share one template, members/index.html, addressed as members/?id=<id> (append &lang=cn for Chinese). It loads data/members/<id>.json and auto-pulls that member's papers from the publications manifest. Personal publication news joins data/news.json by publication_id and is restricted to verified first/co-first and corresponding/co-corresponding author_roles. Legacy stories with the same paper ID are replaced; personal milestones and an expandable archive remain.
• Publications live as one JSON per paper under data/publications/. manifest.json is generated by tools/build-pubs.py (validates required fields, author IDs against team.json, and dedupes). CI does this for you: PRs are validated, and the manifest is rebuilt + committed automatically on merge — see .github/workflows/. Running the script locally is optional, for preview.
• Author names in publication lists are auto-linked to member pages, with the current member highlighted on their own profile (hyphen-tolerant matching, e.g. "Bicong Yan" ↔ "Bi-Cong Yan").

👥 Updating team members

PI edits data/team.json, adding an entry under the relevant category: leader, pi, postdoc, phd_engineering, phd_medical, master_engineering, master_medical, honors_students, alumni.

Members without a photo get an initials-avatar fallback automatically. Set "page": true to give a member a personal page.

Members can update their own page — see MEMBER_GUIDE.md (bilingual, human- and agent-friendly). It covers updating your profile, photo, and publications, and the Pull-Request → owner-review-and-merge workflow. In short: copy data/members/_template.json to data/members/<your-id>.json, fill it in, drop a photo in images/members/, preview locally, and open a Pull Request.

🚀 Local preview

python3 -m http.server 8000

Then open http://localhost:8000.

📦 Deployment

Deployed automatically by GitHub Pages from the main branch. Push to main and the live site updates within ~1 minute.

📄 License

MIT — the website code. Member photos, paper figures, and logos remain the property of their respective owners.

---

AI4Rad Lab · Department of Radiology · Shanghai Sixth People's Hospital · Shanghai Jiao Tong University School of Medicine

🌐 ai4rad-ssph.github.io · 🐙 GitHub · 📧 zhangrp@sjtu.edu.cn