A Computational Pipeline for Advanced Analysis of 4D Flow MRI in the Left Atrium

nnU-Net Models for 4D Flow MRI Cardiac Segmentation

Pre-trained nnU-Net models for automated segmentation of cardiac structures in 4D Flow MRI Phase Contrast Magnetic Resonance Angiography (PC-MRA) images from the pre-print [A Computational Pipeline for Advanced Analysis of 4D Flow MRI in the Left Atrium](https://arxiv.org/abs/2505.09746). These models were trained as part of a comprehensive pipeline for processing and analyzing 4D Flow MRI data.

Models Included

- Left Atrium (la): Segmentation of left atrial chamber
- Left Ventricle (lv): Segmentation of left ventricular chamber  
- Ventricular Outflow (vo): Segmentation of ventricular outflow tract and whoel left heart

Technical Specifications

- Architecture: nnU-Net v2 (3D full resolution configuration)
- Input: Time-averaged PC-MRA images derived from 4D Flow MRI
- Output: Binary segmentation masks in NIfTI format
- Training Data: Multi-center 4D Flow MRI datasets with expert manual annotations

Applications

These models enable automated cardiac structure segmentation for:
- Hemodynamic analysis of blood flow patterns
- Viscous energy loss quantification
- 3D visualization and computational fluid dynamics studies
- Clinical research in cardiovascular imaging

Usage

Models are compatible with nnU-Net v2 framework and can be integrated into the accompanying 4D Flow MRI processing pipeline for end-to-end analysis from raw DICOM data to quantitative flow metrics.

Citation

Please cite the original nnU-Net paper when using these models:
Isensee, F., Jaeger, P. F., Kohl, S. A., Petersen, J., & Maier-Hein, K. H. (2021). nnU-Net: a self-configuring 
method for deep learning-based biomedical image segmentation. Nature methods, 18(2), 203-211.