metadata

title: README
emoji: 🔥
colorFrom: purple
colorTo: pink
sdk: static
pinned: false

Lighter zoo x CT-FM: Through lighter zoo we provide several models pre-trained using the CT-FM vision foundation model for Computed Tomography (CT) scans.

CT-FM is a large-scale 3D image-based pre-trained model designed for diverse radiological tasks. The model was pre-trained on 148,000 CT scans from the Imaging Data Commons using label-agnostic contrastive learning.

Model Details

The model demonstrates strong capabilities across multiple tasks:

Whole-body multi-structure segmentation
Heterogenous tumor segmentation across 4 anatomical sites
Head CT triage
Medical image retrieval
Semantic understanding of anatomical structures Key features:
Learns anatomical clustering without explicit labels
Identifies similar anatomical structures across different scans
Shows robustness in test-retest scenarios
Provides interpretable salient regions in its embeddings

Models Available

Feature extractor ct_fm_feature_extractor which can be used for several feature-based tasks such as image retrieval, semantic search and outlier detection
Fine-tuned whole body segmentation model whole_body_segmentation that segments 117 labels from the TotalSegmentator dataset

Installation

We provide pre-trained as well as fine-tuned models in the lighter-zoo package that interfaces with HF to provide easy to use APIs

To install the lighter-zoo package, use pip:

pip install lighter-zoo

Inspect specific models to see how you can interact with these