SeqSeg: Sequential Vessel Segmentation and Tracking

Automatic tracking and segmentation of blood vessels in CT and MR images using deep learning and geometric tracking.

Why SeqSeg?

SeqSeg segments vessels sequentially, taking steps along vessel centerlines and detecting bifurcations to grow complete vascular trees from just 1–2 seed points. By combining local deep-learning predictions (nnU-Net) with geometric tracking, it stays robust across vessel scales — from small coronaries to large aortas.

• 🌱 Minimal supervision — only 1–2 seed points to initialize
• 🌿 Robust bifurcation detection — automatically follows every branch
• 🩻 Multi-modal — works with CT and MR 3D medical images
• 📏 Scalable — vessels from ~1mm coronaries to ~30mm aortas
• ✅ Clinically validated — coronary, aortic, cerebral, and pulmonary anatomies (pre-trained weights for aorta CT/MR and coronary CT)
• ⚡ Fast — on CPU, about 2–5 minutes for an aorta, 5–15 for a coronary tree, and 10–30 for a cerebral tree. GPU is faster. Details in Benchmarks.

A finished case includes a simvascular/ project you can open directly in SimVascular, with pathlines and contours for every segmented branch.

Real-time demonstration: automatic segmentation of an abdominal aorta in a 3D MR scan.

Start here

• First segmentation — the tutorial runs a sample abdominal-aorta MR scan.
• Your own images — Usage describes images/ and seeds.json. The command is in Quick Start.
• Train a model — Training.

Quick Start

Python 3.9 or newer. 3.11 is the version used in the tutorial and conda example.

The tutorial is the sample run (abdominal-aorta MR, seed points, and Windows notes). Plan on 15–20 minutes including setup. Tracing is capped, so that part finishes in a few minutes; a real case should use the defaults.

Install SeqSeg and the aorta weights below, then run seqseg run batch on your own data. The folder needs images/ and seeds.json (Usage). This example is an aorta MR case. Change -train_dataset, -config_name, and -img_ext to match your images and weights.

pip install seqseg

# Aorta and femoral weights (MR and CT)
curl -L -o nnUNet_results.zip https://zenodo.org/records/15020477/files/nnUNet_results.zip
unzip nnUNet_results.zip

Aorta and femoral weights (nnUNet_results.zip). Coronary CT weights are a separate archive, nnUNet_results_coronary.zip.

-nnunet_results_path is the folder that contains Dataset005_SEQAORTANDFEMOMR (and the other dataset folders in that zip).

seqseg run batch \
    -data_dir your_data/ \
    -nnunet_results_path nnUNet_results/ \
    -train_dataset Dataset005_SEQAORTANDFEMOMR \
    -config_name global_aorta \
    -img_ext .mha \
    -outdir results/

Anatomy	-train_dataset	-config_name
Aorta / femoral MR	Dataset005_SEQAORTANDFEMOMR	global_aorta
Aorta / femoral CT	Dataset006_SEQAORTANDFEMOCT	global_aorta
Coronary CT	Dataset010_SEQCOROASOCACT	global_coro

The tutorial uses -config_name aorta_tutorial for its sample scan. The trainer-folder layout is in Installation. The full command list and flag spelling are in Usage.

Documentation

Job	Guide	Description
Run	Installation	Setup, dependencies, and pre-trained model weights
Run	Tutorial	End-to-end aorta example with sample data
Run	Usage	Data layout, seeds, CLI arguments, and output files
Run	Configuration	YAML configs and tracking parameters
Train	Training	Train nnU-Net models on a new dataset
Understand	Algorithm Overview	Methodology, workflow, and training strategy
Understand	Performance & Benchmarks	Accuracy, timing, and qualitative comparisons
Integrate	API	Call tracing from Python
Integrate	Research & Development	SimVascular, 3D Slicer, and repository layout

Citation

When using SeqSeg, please cite the following paper:

@Article{SveinssonCepero2024,
author={Sveinsson Cepero, Numi
and Shadden, Shawn C.},
title={SeqSeg: Learning Local Segments for Automatic Vascular Model Construction},
journal={Annals of Biomedical Engineering},
year={2024},
month={Sep},
day={18},
issn={1573-9686},
doi={10.1007/s10439-024-03611-z},
url={https://doi.org/10.1007/s10439-024-03611-z},
}