Related Experiment Video
Updated: Dec 30, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.3K
Improving cardiac MRI convolutional neural network segmentation on small training datasets and dataset shift: A
Fumin Guo1, Matthew Ng1, Maged Goubran1
1Sunnybrook Research Institute, University of Toronto, Toronto M4N 3M5, Canada; Department of Medical Biophysics, University of Toronto, Toronto, Canada.
Medical Image Analysis
|January 24, 2020
Summary
This study enhances cardiac MRI segmentation using a novel deep learning and interpretable machine learning approach. The method improves accuracy and reproducibility, even with limited training data for cardiovascular disease analysis.
Area of Science:
- Medical Imaging
- Cardiovascular Disease
- Artificial Intelligence in Medicine
Background:
- Cardiac magnetic resonance imaging (MRI) is crucial for cardiovascular disease care, requiring accurate segmentation of cardiac structures.
- Deep convolutional neural networks (CNNs) excel at image segmentation but often need extensive training data and yield suboptimal results.
Purpose of the Study:
- To enhance multi-class cardiac MRI segmentation by integrating CNNs with interpretable machine learning.
- To improve segmentation accuracy and reduce variability, especially when using limited training datasets.
Main Methods:
- Developed a continuous kernel cut segmentation algorithm combining normalized cuts and continuous regularization.
- Solved the formulation using upper bound relaxation and a continuous max-flow algorithm, with CNN predictions as input.
- Applied and evaluated the approach on diverse cardiac MRI datasets with various cardiovascular pathologies.
Main Results:
- Significantly improved baseline CNN segmentation performance across different datasets.
- Substantially reduced segmentation variability compared to standard CNN approaches.
- Achieved excellent segmentation accuracy with minimal additional computational cost.
Conclusions:
- The proposed method enhances CNN applicability for cardiac MRI segmentation, particularly with small training datasets.
- Improves segmentation accuracy and reproducibility for research and clinical patient care in cardiovascular imaging.
- Offers a robust solution for biomarker enumeration in cardiovascular disease management.
