Related Experiment Video
Updated: Jun 12, 2026
06:45
Automated Joint Space Detection Improves Bone Segmentation Accuracy
Published on: November 28, 2025
Impact of data augmentation size on deep learning-based third lumbar vertebra computed tomography skeletal muscle
Xuzhi Zhao1, Yi Du2,3,4, Xianggao Zhu2,4
1School of Information Network Security, People's Public Security University of China, Beijing, China.
Quantitative Imaging in Medicine and Surgery
|June 11, 2026
Summary
Increasing data augmentation size improves deep learning models for skeletal muscle segmentation on L3 CT scans. An augmentation factor of 6x or more per image is recommended for optimal accuracy and generalization.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Radiology
Background:
- Deep learning (DL) models are used for skeletal muscle segmentation in L3 CT images.
- Data augmentation is common but its impact on segmentation performance is not well understood.
Purpose of the Study:
- To quantitatively assess the effect of augmentation size on DL model performance for L3 skeletal muscle segmentation.
- To provide evidence-based recommendations for optimal augmentation strategies.
Main Methods:
- 400 patient CT scans were used, with L3 skeletal muscle manually annotated.
- Three DL models (U-Net, attention U-Net, attention V-Net) were trained with varying augmentation sizes (1x to 9x augmented images per original).
- Performance was evaluated using Dice Similarity Coefficient (DSC), precision, recall, Hausdorff distance (HD95), and average surface distance (ASD).
Main Results:
- Model performance generally improved with increased augmentation.
- All models performed best with 6x or greater augmentation.
- Attention V-Net achieved the highest performance across metrics, with DSC of 97.627% and ASD of 0.250 mm.
Conclusions:
- Data augmentation significantly enhances DL model performance for L3 CT skeletal muscle segmentation.
- An augmentation size of 6x or greater per image is recommended for high accuracy and generalization.
- This study provides guidance for optimizing augmentation in medical image segmentation tasks.
