Related Experiment Video
Updated: Jun 19, 2026
10:14
3D Scanning Technology Bridging Microcircuits and Macroscale Brain Images in 3D Novel Embedding Overlapping Protocol
Published on: May 12, 2019
An evolutionary neural architecture search for magnetic resonance image reconstructions
Samira Vafay Eslahi1,2, Jian Tao3,4, Jim Ji5,6
1Department of Electrical and Computer Engineering, Texas A&M University, College Station, TX, USA. vafayeslahisamira@gmail.com.
Summary
Automated design of convolutional neural networks (CNNs) using genetic algorithms (GAs) significantly enhances magnetic resonance imaging (MRI) reconstruction accuracy. This approach eliminates manual tuning, offering an efficient solution for medical imaging deep learning models.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Biomedical Engineering
Background:
- Deep learning shows promise for magnetic resonance imaging (MRI) reconstruction from undersampled data.
- Manual design of neural network architectures for MRI reconstruction is inefficient and may not achieve optimal accuracy.
Purpose of the Study:
- To propose an automatic method for designing convolutional neural network (CNN) models optimized for MRI reconstruction accuracy.
Main Methods:
- Introduced a genetic algorithm (GA) framework to optimize CNN architectures.
- Tuned fractional representations of architectural components instead of conventional hyperparameters.
- Explored the design space flexibly for efficient identification of optimal architectures.
Main Results:
- The GA-based automatically designed network achieved higher accuracy than manually designed CNNs on the fastMRI dataset.
- Outperformed existing architectures on brain MRI reconstructions.
- Demonstrated adaptability and generalization across different anatomy types, including knee MRI.
Conclusions:
- Automated CNN design using GA significantly improves MRI reconstruction quality.
- Eliminates the need for manual architecture engineering in deep learning models for medical imaging.
- Offers an efficient and scalable solution for high-performance medical imaging applications.
