Related Experiment Video
Updated: Jun 5, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
368
A lightweight adaptive spatial channel attention efficient net B3 based generative adversarial network approach for
Penta Anil Kumar1, Ramalingam Gunasundari1
1Dept. of Electronics and Communication Engineering, Puducherry Technological University, Puducherry 605014, India.
Magnetic Resonance Imaging
|December 13, 2024
Summary
This study introduces a novel deep learning model, ASCA-EffNet GAN, for faster and higher-quality Magnetic Resonance Imaging (MRI) reconstruction. The new method significantly improves image quality from undersampled data, showing promise for clinical use.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Biomedical Engineering
Background:
- Magnetic Resonance Imaging (MRI) is crucial for non-invasive medical diagnosis due to its soft tissue contrast.
- Traditional MRI acquisition is slow, limiting its clinical efficiency.
- Compressed Sensing MRI (CS-MRI) accelerates acquisition but faces challenges with artifacts and slow iterations, especially at high acceleration factors.
Purpose of the Study:
- To develop a fast and high-quality MR image reconstruction method for CS-MRI.
- To address the limitations of conventional CS-MRI, including sluggish iterations and artifacts.
- To leverage deep learning for improved image reconstruction from undersampled k-space data.
Main Methods:
- Proposed a lightweight Adaptive Spatial-Channel Attention EfficientNet B3-based Generative Adversarial Network (ASCA-EffNet GAN).
- Employed a U-net generator with EfficientNet B3 encoder blocks and a ResNet decoder.
- Utilized a binary classifier discriminator incorporating EfficientNet B3 and adaptive attention mechanisms for spatial and channel-wise feature capture.
Main Results:
- ASCA-EffNet GAN demonstrated superior performance across various metrics compared to conventional reconstruction methods.
- The model achieved remarkable reconstruction capabilities even under high undersampling rates.
- EfficientNet B3's compound scaling balanced model depth, width, and resolution, reducing parameters while optimizing performance.
Conclusions:
- ASCA-EffNet GAN offers a promising solution for accelerating MRI acquisition while maintaining high image quality.
- The adaptive attention mechanisms contribute to detailed anatomical structure reconstruction.
- The proposed method is suitable for clinical applications due to its efficiency and effectiveness in fast MRI.
Keywords:
Adaptive spatial-channel attention(ASCA) mechanismEfficient netGenerative adversarial networks (GANs)Image reconstructionMagnetic resonance imaging (MRI)
