Related Experiment Video
Updated: May 27, 2025
07:53
Author Spotlight: Advancing 3D Modeling for Enhanced Diagnosis and Treatment of Pulmonary Nodules in Early-Stage Lung Cancer
Published on: October 13, 2023
1.3K
A cross-type multi-dimensional network based on feature enhancement and triple interactive attention for LDCT
Lina Jia1,2, Beibei Jia1, Zongyang Li1
1School of Physics and Electronic Engineering, Shanxi University, Taiyuan, China.
Journal of X-Ray Science and Technology
|February 20, 2025
Summary
This study introduces an enhanced deep learning network for low-dose computed tomography (LDCT) image denoising, significantly improving image quality and detail preservation. The new method achieves superior performance in abdominal CT scans compared to existing advanced algorithms.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Image Processing
Background:
- Deep learning methods for low-dose computed tomography (LDCT) image denoising show promise but struggle with structure loss and efficiency.
- Existing techniques often fail to preserve crucial edge information and can be inefficient.
Purpose of the Study:
- To enhance the denoising quality of LDCT images.
- To introduce an improved multi-dimensional hybrid attention network incorporating edge detection for superior LDCT image denoising.
Main Methods:
- Developed an edge enhancement module using trainable Sobel convolution.
- Integrated an enhanced triplet attention network (ETAN) after convolutional layers for comprehensive feature extraction and noise suppression.
- Employed a combined loss strategy (Total Variation Loss + Mean Squared Error) to minimize artifacts and balance denoising with detail preservation.
Main Results:
- The proposed model achieved superior Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) values for abdominal CT images.
- Achieved PSNR of 34.8211 and SSIM of 0.9131, outperforming advanced algorithms like CT-former, REDCNN, and EDCNN.
- Demonstrated significant improvements in both subjective visual quality and objective performance metrics.
Conclusions:
- The enhanced network effectively improves LDCT image denoising quality.
- The proposed algorithm offers significant advancements in visual fidelity and objective performance for LDCT image reconstruction.
- The integration of edge detection and attention mechanisms provides a robust solution for LDCT denoising challenges.
