Related Experiment Video
Updated: Jul 20, 2025
06:48
Automated Segmentation of Cortical Grey Matter from T1-Weighted MRI Images
Published on: January 7, 2019
8.9K
Optimizing Deep Learning for Cardiac MRI Segmentation: The Impact of Automated Slice Range Classification.
Sarv Priya1, Durjoy D Dhruba2, Sarah S Perry3
1Department of Radiology, University of Iowa Carver College of Medicine, Iowa City, Iowa (S.P.).
Academic Radiology
|August 4, 2023
Summary
Automated deep learning (DL) cardiac segmentation is improved by adding an initial slice range classification step. This integration enhances accuracy and reduces observer bias in cardiovascular MRI analysis.
Area of Science:
- Cardiovascular Imaging
- Artificial Intelligence in Medicine
- Medical Image Analysis
Background:
- Cardiac magnetic resonance imaging (CMR) is vital for cardiovascular disease diagnosis.
- Manual segmentation in CMR is time-consuming and prone to observer variability.
- Deep learning (DL) offers automated segmentation but is limited by slice selection.
Purpose of the Study:
- To enhance DL-based cardiac segmentation by integrating an automated slice range classification step.
- To evaluate the performance of a combined classification and segmentation model against a stand-alone segmentation model.
- To compare automated segmentation results with manual segmentation using established metrics.
Main Methods:
- Utilized a public cardiac MRI dataset (Multi-Disease, Multi-View & Multi-Center Right Ventricular Segmentation) with short-axis cine data.
- Developed and tested three classification and seven segmentation DL models.
- Assessed the top segmentation model with and without an integrated classification step (CBAM-integrated 2D-CNN and 2D-UNet).
- Validated models using Dice score, Hausdorff distance, correlation, and Bland-Altman plots.
Main Results:
- The combined classification and segmentation model achieved superior Dice scores: 0.952 (LV), 0.933 (RV), and 0.875 (myocardium).
- High correlation (0.92-0.99) was observed between automated and manual segmentation for biventricular volumes, ejection fraction, and myocardial mass.
- Mean absolute differences for clinical parameters were within interobserver variability, demonstrating comparable performance to manual annotation.
Conclusions:
- Integrating automated slice range classification significantly improves DL-based cardiac chamber segmentation performance.
- The proposed method offers a more accurate and reliable automated solution for cardiac MRI analysis.
- This approach has the potential to reduce postprocessing time and observer bias in clinical practice.
