Skip to content

FMS-Medical: Foundation Models for Advancing Healthcare (Catalog)

Overview

Type: Knowledge Base / Survey Resource
Format: Curated repository of medical foundation models + datasets
Coverage: Language (LFM), Vision (VFM), Bioinformatics (BFM), Multimodal (MFM)
Primary use: Model and dataset discovery, benchmarking reference, literature review

Purpose & Design Philosophy

FMS-Medical is an "awesome list" style knowledge base tracking foundation model research across healthcare modalities. Maintained as a GitHub repository with bilingual documentation (English + Chinese), it provides structured references to 100+ medical FM papers, datasets, tutorials, and related resources organized by modality and year. The resource is anchored by an IEEE Reviews in Biomedical Engineering survey paper and serves as a comprehensive entry point for medical AI research.

Key value: Centralized, actively maintained catalog of medical FMs and datasets—ideal for systematic literature review and baseline selection.

Catalog Structure

Model Taxonomies

Category Coverage Use Case
LFM (Language) Medical LLMs, clinical NLP Text understanding, report generation, QA
VFM (Vision) Medical image encoders, segmentation Radiology, pathology, ultrasound analysis
BFM (Bioinformatics) Genomics, proteomics, drug discovery Sequence modeling, variant interpretation
MFM (Multimodal) Vision-language, integrated models Unified diagnosis, multimodal reasoning

Dataset Catalogs

  • Text datasets: Clinical notes, radiology reports, biomedical literature
  • Imaging datasets: CXR, CT, MRI, pathology, ultrasound, retinal
  • Omics datasets: Genomics, transcriptomics, proteomics
  • Multimodal datasets: Image-text pairs, integrated EHR + imaging

Integration Strategy

For Neuro-Omics KB

FMS-Medical provides dataset discovery and model benchmarking:

Key uses: - Literature review: Identify related work in medical AI for neuro-omics - Dataset selection: Find imaging and genomics datasets for validation - Baseline comparison: Track state-of-the-art methods for benchmarking - Survey reference: Cite comprehensive medical FM survey

Application to KB pipeline: 

FMS-Medical catalog
    ↓
Extract relevant entries:
    - Brain imaging models
    - Genomics foundation models
    - Multimodal medical datasets
    ↓
Populate KB model/dataset cards
    ↓
Benchmark neuro-omics methods against medical FM baselines

For ARPA-H Brain-Omics Model (BOM)

FMS-Medical informs medical AI landscape understanding:

Survey medical FMs
    ↓
Identify integration patterns:
    - CLIP-style alignment (M3FM, TITAN)
    - MoT/MoE architectures
    - LLM continual pretraining (Me-LLaMA)
    ↓
Apply to BOM design

Transfer insights: - Comprehensive coverage: Survey spans all medical FM modalities—identify gaps for neuro-omics - Dataset catalogs: Find publicly available datasets for cross-validation - Benchmark references: Track medical FM performance for comparison - Bilingual support: Chinese documentation aids international collaboration

Catalog Access Workflow

# 1. Clone or sync FMS-Medical repository
git clone https://github.com/YutingHe-list/Awesome-Foundation-Models-for-Advancing-Healthcare

# 2. Review README for model/dataset tables

# 3. Extract entries relevant to neuro-omics:
#    - VFM: Brain imaging models
#    - BFM: Genomics models
#    - MFM: Multimodal integration patterns

# 4. Populate KB YAML cards from extracted entries

# 5. Track updates via GitHub (actively maintained)

For KB automation: - Parse README tables to generate candidate model/dataset cards - Link to survey PDFs for detailed descriptions - Monitor repository updates for new medical FMs

Strengths & Limitations

Strengths

  • Comprehensive coverage: 100+ medical FMs across all modalities
  • Actively maintained: Regular updates with publication news
  • Bilingual: English + Chinese documentation
  • Well-organized: Structured by modality, year, and venue
  • Survey anchor: Peer-reviewed IEEE Reviews paper provides synthesis

Limitations

  • Documentation-only: No executable code or model weights
  • General medical focus: Limited neuro-omics specific content
  • No unified schema: Markdown tables, not structured YAML/JSON
  • Citation lag: New models may take time to appear in catalog

When to Use FMS-Medical

✅ Use when: - Starting literature review on medical FMs - Selecting baseline models for benchmarking - Finding publicly available medical datasets - Identifying integration patterns from related work - Citing comprehensive medical FM surveys

⚠️ Not a substitute for: - Model implementation code (links to external repos) - Pretrained model weights (links to paper/HuggingFace) - Executable benchmarking pipelines (reference only)

⚠️ Complement with: - Papers With Code: For benchmark leaderboards - HuggingFace Model Hub: For pretrained weights - Model-specific repos: For implementation details

Reference Materials

Knowledge Base Resources

Curated materials in this KB: - Code walkthrough: FMS-Medical walkthrough - Dataset card (YAML): kb/datasets/fms_medical_catalog.yaml

Integration recipes: - Integration Strategy — Dataset selection - KB Overview — Catalog integration patterns

Original Sources

Source repositories: - Local copy: external_repos/fms-medical/ - Official GitHub: YutingHe-list/Awesome-Foundation-Models-for-Advancing-Healthcare

Original survey paper: - Title: "Foundation Models for Healthcare" - Authors: Yuting He, et al. - Published: IEEE Reviews in Biomedical Engineering, 2024 - Link (IEEE): IEEE Xplore: 10750441 - Link (arXiv): arXiv:2404.03264

Next Steps in Our Pipeline

  1. Systematic extraction: Parse FMS-Medical tables to populate KB model/dataset cards
  2. Gap analysis: Identify medical FM capabilities missing from neuro-omics KB
  3. Benchmark selection: Choose medical FM baselines for comparison experiments
  4. Dataset discovery: Find publicly available datasets for cross-validation
  5. Literature tracking: Monitor repository updates for new medical FM methods

Engineering Notes

  • FMS-Medical is pure documentation—no code dependencies
  • Bilingual PDFs in files/ directory useful for international teams
  • Citation metadata in tables ready for automated KB card generation
  • Active maintenance—check GitHub for recent updates before citing
  • Survey paper provides narrative synthesis complementing tabular catalog