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
Data Representations for Segmentation of Vascular Structures Using Convolutional Neural Networks with U-Net
Summary
Convolutional neural networks (CNNs) show improved medical image segmentation for thin, curved structures like coronary arteries. A 3D cross-sectional data representation significantly enhances segmentation performance in computed tomography angiography (CTA).
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Cardiovascular Imaging
Background:
- Convolutional neural networks (CNNs) are effective for medical image segmentation.
- Segmenting small, curved structures like coronary arteries in computed tomography angiography (CTA) presents significant challenges.
Purpose of the Study:
- To evaluate the impact of data representation on the performance of U-Net architecture CNNs for segmenting tubular structures.
- To compare 2D and 3D data representations in cross-sectional and Cartesian formats.
Main Methods:
- Utilized U-Net architecture convolutional neural networks (CNNs).
- Investigated 2D and 3D input data using cross-sectional and Cartesian representations.
- Assessed segmentation performance using Dice coefficients and Hausdorff distances.
Main Results:
- Data representation significantly impacted segmentation outcomes, with Dice coefficients ranging from 60% to 82%.
- Hausdorff distances varied between 1.38 mm and 5.90 mm.
- Achieved results comparable to human expert annotations.
Conclusions:
- A 3D cross-sectional data representation is optimal for segmenting thin, tubular structures.
- The choice of data representation is crucial for improving CNN-based medical image segmentation accuracy.
