Skip to content

Hugging Face Paper

CT-FM: A 3D Image-Based Foundation Model for Radiological Tasks

CT-FM Overview

Brief

This repository contains the code and resources for CT-FM, a 3D image-based pre-trained foundation model designed for various radiological tasks. CT-FM is trained using self-supervised learning (SSL) on a large dataset of 148,000 CT scans. This model aims to address a range of tasks, including whole-body segmentation, tumor segmentation, head CT triage, and medical image retrieval. This work builds upon previous efforts in radiological AI, shifting from task-specific expert models to unified foundation models for broader adaptability and efficiency.

Key Innovations

  • Large-Scale 3D Pretraining: Emphasis on 3D data rather than traditional 2D datasets.
  • Task-Agnostic Training: Enabling transferability across various radiological tasks.
  • Open Source: Model weights, data, and code are shared for collaborative development.



  • Downloading Data

    All datasets used in the study are public

    Download data

  • Use CT-FM models

    CT-FM feature extractors and trained downstream models are available on HF

    Go to HF

  • Reproduce our pre-training framework

    Implement our pre-training method on your own data

    Pretraining instructions

  • Build your projects using Lighter

    CT-FM ♥ Lighter
    All pre-training is performed using lighter Explore here