Awesome
Compositional Segmentation of Cardiac Images Leveraging Metadata
Accepted at WACV 2025 (IEEE/CVF Winter Conference on Applications of Computer Vision)
You can view or download the paper HERE.
Overview:
This repository provides the code implementation for our paper, "Compositional Segmentation of Cardiac Images Leveraging Metadata." Our work presents a novel compositional segmentation approach to efficiently segment cardiac structures by integrating metadata, such as patient demographics and acquisition parameters, to modulate the segmentation network conditionally.
Key Contributions:
Compositional Segmentation: A hierarchical approach that performs super and sub-segmentation.
Super-segmentation: Localizes the heart.
Sub-segmentation: Further segments detail structures within the heart, including the left and right ventricles (LV, RV) and the myocardium (MYO).
Cross-Modal Feature Integration (CMFI): We introduce a CMFI module to leverage metadata (e.g., scanner type, medical condition, demographic details) as additional context, enhancing segmentation accuracy and robustness.
Evaluation:
Our approach was evaluated on two modalities using publicly available datasets:
M&Ms-2 Dataset (MRI)
CAMUS Dataset (Ultrasound)
Results demonstrate state-of-the-art segmentation performance across diverse cardiac imaging modalities.
Training Steps
Segmentation Model
The segmentaton models do not apply any activation on logits, so add activation after the last 1x1 conv layer or incorporate it in your loss function. Change Num_Classes to your (Segmentation class + 1).(1 for background). We have provided both versions
Citation
@inproceedings{author2025compsegmetadata,
title={Compositional Segmentation of Cardiac Images Leveraging Metadata},
author={Abbas Khan, Muhammad Asad, Martin Benning, Caroline Roney, and Greg Slabaugh},
booktitle={Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
year={2025} }