Ultra-Fast T2-Weighted MR Reconstruction Using Complementary T1-Weighted Information.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Routine Magnetic Resonance Imaging (MRI) protocols like T1-weighted imaging (T1WI) and T2-weighted imaging (T2WI) offer complementary diagnostic information but suffer from long acquisition times (~10 min).
• Long MRI acquisition times increase vulnerability to motion artifacts, compromising image quality.
• Existing accelerated MRI reconstruction algorithms typically utilize data from a single imaging protocol, limiting their potential.

Purpose of the Study:

• To develop a novel method for accelerating MRI acquisition by leveraging information from multiple complementary protocols.
• To reconstruct high-quality T2-weighted images (T2WI) from under-sampled data by combining T1-weighted imaging (T1WI) information.
• To introduce a new deep learning model, Dense-Unet, for efficient and high-performance MRI reconstruction.

Main Methods:

• A novel deep learning architecture, Dense-Unet, was developed for MRI reconstruction.
• The approach integrates complementary MRI protocols, specifically using T1WI data to aid the reconstruction of under-sampled T2WI.
• The Dense-Unet model was trained and evaluated for its ability to reconstruct fully-sampled 3D T2WI from accelerated data.

Main Results:

• The proposed Dense-Unet model successfully reconstructed 3D T2WI volumes in under 10 seconds, achieving acceleration rates of 8x or higher.
• Reconstructed images exhibited negligible aliasing artifacts and minimal signal-noise-ratio (SNR) loss compared to fully-sampled acquisitions.
• Dense-Unet demonstrated superior performance with fewer parameters and reduced computational requirements compared to other methods.

Conclusions:

• Combining complementary MRI protocols (T1WI and T2WI) is a viable strategy for accelerating MRI acquisition.
• The novel Dense-Unet deep learning approach enables rapid, high-quality MRI reconstruction with significant acceleration.
• This work represents the first utilization of multi-protocol data for accelerating target MRI sequence reconstruction, offering a promising direction for clinical practice.