Related Experiment Video
Updated: Sep 21, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.0K
Deep Neural Network for Cardiac Magnetic Resonance Image Segmentation
David Chen1, Huzefa Bhopalwala2, Nakeya Dewaswala2
1Department of Cardiovascular Surgery, Cleveland Clinic, Cleveland, OH 44195, USA.
Journal of Imaging
|May 27, 2022
Summary
Automated cardiac magnetic resonance (CMR) image segmentation is faster with a new multi-image type UNet (MI-UNet). This AI model improves segmentation accuracy for left ventricular structures in multiple CMR image types, aiding clinical analysis.
Area of Science:
- Cardiovascular Imaging
- Artificial Intelligence in Medicine
- Medical Image Analysis
Background:
- Cardiac magnetic resonance (CMR) image analysis is time-consuming.
- Automated segmentation of cardiac structures can significantly reduce analysis time.
- Leveraging spatial similarities across different CMR image types can enhance segmentation accuracy.
Purpose of the Study:
- To develop and evaluate a multi-image type UNet (MI-UNet) model for joint segmentation of multiple CMR image sequences.
- To compare the segmentation performance of MI-UNet against single-image type UNet models.
- To assess the impact of MI-UNet on the accuracy of cardiac structure segmentation in hypertrophic cardiomyopathy patients.
Main Methods:
- A multi-image type UNet (MI-UNet) model was developed using data from 72 patient exams.
- The model was trained to jointly segment multiple CMR image types, including steady-state free precession (SSFP) and late gadolinium enhancement (LGE).
- Segmentation accuracy was quantified using the Dice similarity coefficient (DSC) and compared between MI-UNet and single-image type UNet models.
Main Results:
- MI-UNet achieved a superior DSC of 0.92 ± 0.06 for SSFP images versus 0.87 ± 0.08 for single-image UNet (p < 0.001).
- MI-UNet achieved a superior DSC of 0.86 ± 0.11 for LGE images versus 0.78 ± 0.11 for single-image UNet (p = 0.001).
- Improved accuracy was most notable for left ventricular structures in SSFP and LGE images; no significant difference was observed for the right ventricle.
Conclusions:
- Joint segmentation of multiple CMR image types using MI-UNet enhances segmentation accuracy for left ventricular structures compared to single-image models.
- The MI-UNet model shows potential to expedite the analysis and interpretation of multi-sequence CMR datasets in clinical practice.
- This AI-driven approach offers a promising solution for improving efficiency in cardiovascular imaging analysis.
