Related Experiment Video
Updated: Nov 2, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
592
Abdominal Aortic Aneurysm Segmentation Using Convolutional Neural Networks Trained with Images Generated with a
Karen López-Linares1,2,3, Maialen Stephens1, Inmaculada García1,2
1Vicomtech Foundation, San Sebastián, Spain.
Machine Learning and Medical Engineering for Cardiovascular Health and Intravascular Imaging and Computer Assisted Stenting : First International Workshop, MLMECH 2019, and 8Th Joint International Workshop, CVII-STENT 2019, Held in Conj
|June 11, 2021
Summary
This study demonstrates that Convolutional Neural Networks (CNNs) can effectively segment abdominal aortic aneurysms (AAAs) using only synthetic images. This approach reduces the need for extensive real-world annotated data, proving comparable to traditional methods.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Computational Biology
Background:
- Abdominal aortic aneurysms (AAAs) are life-threatening vascular conditions requiring accurate imaging for management.
- Computed Tomography Angiography (CTA) is crucial for AAA assessment, with segmentation vital for risk evaluation.
- Current deep learning segmentation methods, like Convolutional Neural Networks (CNNs), demand large annotated datasets, posing a significant challenge.
Purpose of the Study:
- To develop and validate a methodology for training CNNs to segment AAAs using solely synthetic data.
- To assess the generalization capability of CNNs trained on synthetic data for segmenting real CTA scans.
- To reduce the dependency on large annotated medical image datasets for AAA segmentation.
Main Methods:
- Generation of synthetic AAA images using a shape model with realistic deformations derived from principal component analysis of registration data.
- Training a CNN model exclusively on these generated synthetic images.
- Evaluating the trained CNN's performance in segmenting AAAs from unseen, original CTA scans.
Main Results:
- CNNs trained with synthetic data achieved segmentation performance comparable to those trained with real annotated images.
- The proposed methodology demonstrated effective generalization, accurately segmenting AAAs in new CTA scans.
- The findings suggest synthetic data can significantly mitigate the need for extensive manual annotation.
Conclusions:
- Training CNNs with synthetic data is a viable and effective strategy for abdominal aortic aneurysm segmentation.
- This approach holds promise for reducing the data acquisition burden in medical image analysis.
- The methodology can potentially be extended to other aneurysm types and medical imaging segmentation tasks.
