Skip to content

FMS-Medical Curation Walkthrough

KB references: Survey digest (pending) · Integration strategy · KB overview · Experiment config stub

Overview

external_repos/fms-medical is an “awesome list” style knowledge base that tracks foundation-model research across healthcare modalities—language (LFM), vision (VFM), bioinformatics (BFM), and multimodal (MFM)—plus dataset catalogs, tutorials, and bilingual survey PDFs. The maintainers keep the README current with publication news (e.g., IEEE Reviews acceptance) and provide both English and Chinese summaries (files/HFM_Chinese.pdf), making it a convenient seed for KB model/dataset cards.^[text title="/Users/allison/Projects/neuro-omics-kb/external_repos/fms-medical/README.md] (lines 1-38)"

At-a-Glance

Architecture Params Context Inputs Key capabilities Repo
Markdown knowledge graph: NEWS → Survey references → modality-specific method lists + dataset tables.^[text title="/Users/allison/Projects/neuro-omics-kb/external_repos/fms-medical/README.md] Organized by year (2020–2024) and modality; each entry stores venue, short description, and code/paper links—ready for transformation into KB YAML cards. Highlights IEEE Reviews 2024 survey + arXiv companions; hosts bilingual PDF in files/ for regional teams.^[5:38:/Users/allison/Projects/neuro-omics-kb/external_repos/fms-medical/README.md] Pure Markdown + embedded images; contributions occur via pull requests (no runtime code). Quick lookup for LFM/VFM/BFM/MFM models, dataset tables (text, imaging, omics, multimodal), lectures, blogs, and related awesome lists.^[39:400:/Users/allison/Projects/neuro-omics-kb/external_repos/fms-medical/README.md] github.com/YutingHe-list/Awesome-Foundation-Models-for-Advancing-Healthcare

Repository Notes

  • Documentation-only. No Python modules or requirements—syncing the README (and optional PDF) is sufficient to integrate the content into KB templates.
  • Bilingual assets. files/HFM_Chinese.pdf mirrors the survey for Mandarin readers; cite it when translating KB summaries.^[text title="/Users/allison/Projects/neuro-omics-kb/external_repos/fms-medical/README.md] (lines 17-34)"
  • Citation-first. Each entry lists papers/code, so KB automation scripts can parse the tables to populate kb/model_cards, kb/datasets, or docs/generated references.

Key Components

Survey Metadata & NEWS Banner

Top-of-file announcements capture publication milestones, acceptance venues, and contact information. These lines can drive KB changelogs or curated timelines.

README header with news updates:

/Users/allison/Projects/neuro-omics-kb/external_repos/fms-medical/README.md (lines 5-34)
[![Awesome](https://awesome.re/badge.svg)](https://awesome.re)

[NEWS.20241115] **Our survey [paper](https://ieeexplore.ieee.org/document/10750441) has been accepted by IEEE Reviews in Biomedical Engineering (IF: 17.2).**

[NEWS.20240405] **The related survey [paper](https://arxiv.org/abs/2404.03264) has been released.**

[NOTE] **If you have any questions, please don't hesitate to [contact us](mailto:yuting.he4@case.edu).**

Modality Method Registries (LFM/VFM/BFM/MFM)

Each section groups methods by year, venue, and modality. Capturing these rows lets the KB auto-generate candidate model cards or integration experiments.

Method listings by modality:

/Users/allison/Projects/neuro-omics-kb/external_repos/fms-medical/README.md (lines 82-210)
## LFM methods
**2024**
- [AAAI] Zhongjing: Enhancing the chinese medical capabilities of large language model through expert feedback and realworld multi-turn dialogue. [[Paper]](https://ojs.aaai.org/index.php/AAAI/article/view/29907) [[Code]](https://github.com/SupritYoung/Zhongjing)
- [NeurIPS] MDAgents: An adaptive collaboration of LLMs for medical decision-making. [[Paper]](https://arxiv.org/abs/2404.15155) [[Code]](https://github.com/mitmedialab/MDAgents)
...
## VFM methods
**2024**
- [arXiv] USFM: A universal ultrasound foundation model generalized to tasks and organs towards label efficient image analysis. [[paper]](https://arxiv.org/html/2401.00153v2)

Dataset Catalogs

Separate tables detail datasets per modality (text, imaging, multimodal). These rows map neatly onto kb/datasets/*.yaml and help ensure coverage across integrative experiments.

Dataset tables by modality:

/Users/allison/Projects/neuro-omics-kb/external_repos/fms-medical/README.md (lines 339-399)
## Datasets
### LFM datasets
|                           Dataset  Name                               | Text Types  |            Scale           |    Task    |                       Link                             |
| :-------------------------------------------------------------------: | :-------: | :------------------------: | :--------: | :----------------------------------------------------: |
|[PubMed](https://pubmed.ncbi.nlm.nih.gov/download/) | Literature | 18B tokens |  Language modeling |[*](https://pubmed.ncbi.nlm.nih.gov/download/)|
|[MedC-I](https://arxiv.org/abs/2304.14454)| Literature | 79.2B tokens |  Dialogue |[*](https://huggingface.co/datasets/axiong/pmc_llama_instructions)|
...
|[CMeKG-8K](https://www.mdpi.com/2078-2489/11/4/186)| Dialogue | 8K instances |  Dialogue |[*](https://github.com/WENGSYX/CMKG)|

Other Resources (Lectures/Blogs/Awesome lists)

The README also aggregates tutorials, blogs, and related awesome repositories under “Other Resources,” which can seed KB “Further reading” sections or onboarding material.

/Users/allison/Projects/neuro-omics-kb/external_repos/fms-medical/README.md (lines 53-74)
- [Other Resources](#other-resources)
  - [Lectures and tutorials](#lectures-and-tutorials)
  - [Blogs](#blogs)
  - [Related awesome repositories](#related-awesome-repositories)

Integration Hooks (Curation ↔ KB)

  • Automate card creation. Parse the Markdown tables (e.g., via pandoc or custom scripts) to prefill kb/model_cards with metadata (venue, year, links), ensuring coverage parity across modalities.
  • Align dataset registries. Map the README dataset entries to kb/datasets/*.yaml and tag each with modality + task so integration plans can quickly reference scale/availability.
  • Leverage bilingual PDFs. When publishing KB summaries for non-English audiences, cite files/HFM_Chinese.pdf to keep translations aligned with the upstream survey and avoid redundant localization work.