Related Experiment Video
Updated: Jul 1, 2026
11:13
Quantification of Mouse Heart Left Ventricular Function, Myocardial Strain, and Hemodynamic Forces by Cardiovascular Magnetic Resonance Imaging
Published on: May 24, 2021
Scalable Left Ventricular ROI Annotation for Stress Perfusion Cardiac MRI using Deep Learning with Visual Refinement
Mahsa Pourhossein Kalashami1, Mohamed Elshibly2, Simran Shergill2
1School of Engineering, University of Leicester, Leicester, UK. mpk20@leicester.ac.uk.
Journal of Imaging Informatics in Medicine
|June 29, 2026
Summary
This study introduces an efficient framework for localizing the left ventricular region in cardiac MRI scans, significantly reducing annotation time and improving AI analysis for cardiovascular magnetic resonance imaging.
Area of Science:
- Cardiovascular Imaging
- Medical Artificial Intelligence
- Image Processing
Background:
- Accurate left ventricular (LV) region localization in stress perfusion cardiovascular magnetic resonance (CMR) is difficult due to low signal-to-noise ratio (SNR) and motion artifacts.
- Efficient region of interest (ROI) localization is crucial for preprocessing, reducing irrelevant data, and enhancing AI-based analysis in CMR.
Purpose of the Study:
- To develop a scalable and annotation-efficient framework for LV-centered ROI localization and preprocessing in low-SNR stress perfusion CMR.
- To improve the efficiency and accuracy of ROI identification for downstream AI applications in cardiac imaging.
Main Methods:
- Utilized a U-Net model pretrained on cine CMR data for initial LV localization proposals on unannotated perfusion CMR data.
- Employed segmentation outputs to define LV-centered circular ROIs and incorporated a graphical user interface (GUI) for rapid annotation and refinement.
- Applied the framework to 798 stress perfusion videos, with a subset manually annotated and reviewed by clinical experts using the GUI.
Main Results:
- The GUI-assisted annotation framework reduced annotation time by an estimated 8-12 times compared to conventional methods (mean 2.5 min/video).
- Fine-tuning the model on reviewed annotations improved the Dice score from 0.87 to 0.90, indicating enhanced segmentation accuracy.
- The framework successfully processed 798 stress perfusion videos, demonstrating scalability for large datasets.
Conclusions:
- The proposed framework offers a scalable and efficient solution for LV-centered ROI annotation and preprocessing in low-quality stress perfusion CMR.
- This approach facilitates improved AI-based analysis by providing accurate and efficiently localized ROIs, addressing key challenges in cardiac MRI.
- The GUI-assisted workflow significantly reduces manual annotation burden, making it practical for clinical application.
