Robust Deep Learning for Myocardial Scar Segmentation in Cardiac MRI with Noisy Labels

📄 This paper has been accepted by MICCAI 2025 — (https://arxiv.org/abs/2506.21151)

Abstract

The accurate segmentation of myocardial scars from cardiac MRI is essential for clinical assessment and treatment planning. In this study, we propose a robust deep-learning pipeline for fully automated myocardial scar detection and segmentation by fine-tuning state-of-the-art models. The method explicitly addresses challenges of label noise from semi-automatic annotations, data heterogeneity, and class imbalance through the use of Kullback-Leibler loss and extensive data augmentation. We evaluate the model's performance on both acute and chronic cases and demonstrate its ability to produce accurate and smooth segmentations despite noisy labels. In particular, our approach outperforms state-of-the-art models like nnU-Net and shows strong generalizability in an out-of-distribution test set, highlighting its robustness across various imaging conditions and clinical tasks. These results establish a reliable foundation for automated myocardial scar quantification and support the broader clinical adoption of deep learning in cardiac imaging. 📢 Accepted at MICCAI 2025

Architecture Overview

Figure 1: YoloSAM pipeline combining YOLO for detection and SAM for precise segmentation

Sample Results

Figure 2: Sample myocardial scar detection and segmentation results on cardiac MRI

Key Features

• Robust to Noisy Labels: Handles semi-automatic annotation noise using Kullback-Leibler loss
• Two-Stage Pipeline: YOLO for efficient detection + SAM for precise segmentation
• Data Augmentation: Extensive augmentation strategies for improved generalization
• Class Imbalance Handling: Specialized loss functions for imbalanced medical data
• Cross-Domain Robustness: Strong performance on out-of-distribution test sets
• Clinical Ready: Outperforms nnU-Net and other state-of-the-art methods

Quick Start

Installation

# Clone the repository
git clone https://github.com/Danialmoa/YoloSAM
cd YoloSAM

# Install dependencies
pip install -e .

# Download pre-trained models
python utils/download_model.py

Inference

from scripts.inference import YoloSAMInference

# Initialize the model
model = YoloSAMInference()

# Run inference on a cardiac MRI image
result = model.predict("path/to/cardiac_mri.png")

Project Structure

YoloSAM/
├── models/ # Core model implementations
│ ├── yolo.py # YOLO detection model
│ ├── sam.py # SAM segmentation model
│ └── init.py
├── scripts/ # Training and inference scripts
│ ├── train_yolo.py # YOLO training script
│ ├── train_sam.py # SAM fine-tuning script
│ ├── inference.py # Inference pipeline
│ └── init.py
├── utils/ # Utility functions
│ ├── config.py # Configuration management
│ ├── dataset.py # Dataset handling
│ ├── loss.py # Custom loss functions (KL loss + Dice + BC)
│ ├── prompt.py # SAM prompting strategies
│ ├── mask_to_yolo.py # Mask to YOLO format conversion
│ ├── z_score_norm.py # Normalization utilities
│ └── download_model.py # Model downloading
├── notebooks/ # Jupyter notebooks
│ └── YoloSAM_Tutorial.ipynb # Complete tutorial
├── sample_data/ # Example data and structure
│ ├── dataset.yaml # Dataset configuration
│ ├── train/ # Training data
│ └── val/ # Validation data
├── checkpoints/ # Model checkpoints
├── README.md              # Project documentation
├── LICENSE                # MIT License
└── setup.py               # Package installation

Data Structure

The project expects data in the following format:

your_dataset/
├── train/
│ ├── images/ # Cardiac MRI images (.png/.jpg)
│ └── masks/ # Segmentation masks (.png)
├── val/
│ ├── images/
│ └── masks/

Models

YOLO Detection Model (models/yolo.py)

• Efficient object detection for myocardial scar localization
• Handles class imbalance through weighted loss functions
• Optimized for medical imaging with custom anchors

SAM Segmentation Model (models/sam.py)

• Fine-tuned Segment Anything Model for precise segmentation
• Custom prompting strategies for cardiac MRI
• Robust to noisy labels using KL divergence loss

Google Colab Tutorial

For easy experimentation and fine-tuning, check out our comprehensive Jupyter notebook:

The notebook includes:

• Step-by-step setup instructions
• Data preparation and augmentation
• Model training and fine-tuning
• Evaluation and visualization
• Results analysis and comparison

Acknowledgments

We thank the following projects and teams for their foundational work:

• Meta AI Research for developing and open-sourcing the Segment Anything Model (SAM)
• Ultralytics for the YOLOv12 object detection framework
• EMIDEC Challenge for providing sample MRI data for myocardial infarct detection and segmentation

This work builds upon these excellent foundations to advance automated myocardial scar segmentation in cardiac MRI.

Citation

If you use this code in your research, please cite our paper:

  title={Robust Deep Learning for Myocardial Scar Segmentation in Cardiac MRI with Noisy Labels},
  author={Moafi, Aida and Moafi, Danial and Mirkes, Evgeny M and McCann, Gerry P and Alatrany, Abbas S and Arnold, Jayanth R and Ghazi, Mostafa Mehdipour},
  journal={arXiv preprint arXiv:2506.21151},
  year={2025}
}

Contact

For questions and collaborations, please contact: [am1392@leicester.ac.uk, d.moafi@student.unisi.it]

License

This project is licensed under the MIT License - see the LICENSE file for details.