Isotropic Brain MRI Reconstruction from Orthogonal Scans Using 3D Convolutional Neural Network.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Isotropic 3D MRI reconstruction is crucial for accurate volumetric analysis.
• Traditional super-resolution (SR) methods face challenges like poor performance, long processing times, and manual feature engineering.
• Convolutional Neural Networks (CNNs) offer powerful automatic feature extraction capabilities.

Purpose of the Study:

• To develop an end-to-end deep learning strategy for isotropic MRI reconstruction.
• To leverage 3D CNNs for enhanced capture of structural features and precise prediction.
• To utilize complementary information from multiple orthogonal scans for improved inference.

Main Methods:

• Implementation of a deep learning strategy based on 3D convolutional neural networks (3D CNNs).
• Inputting multiple orthogonal MRI scans to exploit complementary dimensional information.
• Training and validation of the 3D CNN model for isotropic volume reconstruction.

Main Results:

• The proposed 3D CNN method demonstrates promising quantitative and qualitative performance.
• Achieved reconstruction of a 256x256x256 3D MRI volume in under 1 minute using a GPU.
• Outperformed traditional SR methods in accuracy and processing speed.

Conclusions:

• The deep learning-based approach provides a superior and practical solution for isotropic MRI reconstruction.
• This method significantly reduces processing time while maintaining high accuracy.
• Enables efficient generation of isotropic 3D MRI volumes for advanced analysis.