Single-shot T₂ mapping using overlapping-echo detachment planar imaging and a deep convolutional neural network.

Area of Science:

• Magnetic Resonance Imaging (MRI)
• Artificial Intelligence in Medical Imaging
• Deep Learning for Image Reconstruction

Background:

• Quantitative magnetic resonance imaging (MRI) provides valuable diagnostic information.
• T2 mapping is crucial for characterizing tissue properties.
• Single-shot Overlapping-Echo Detachment (OLED) planar imaging offers fast acquisition but requires robust reconstruction methods.

Purpose of the Study:

• To develop and validate an end-to-end deep convolutional neural network (CNN) for efficient and reliable T2 mapping reconstruction.
• To utilize a deep residual network (ResNet) architecture for improved T2 mapping from OLED sequences.
• To demonstrate the generalization capability of the CNN from simulated to in vivo human brain data.

Main Methods:

• A deep residual network (ResNet) was trained using simulated data generated by SPROM software.
• The ResNet learned the complex relationship between OLED images and T2 maps.
• The trained network was applied to reconstruct T2 maps from both simulated and in vivo human brain datasets.

Main Results:

• The ResNet, trained solely on simulated data, demonstrated excellent generalization to real human brain data.
• The proposed CNN-based method significantly outperformed the conventional echo-detachment-based reconstruction.
• Highly accurate T2 mapping was achieved in approximately 30 milliseconds, a substantial improvement over the 2-minute traditional method.

Conclusions:

• The developed deep learning approach enables efficient and accurate T2 mapping from OLED sequences.
• This method facilitates real-time dynamic and quantitative MRI.
• Deep convolutional neural networks show significant potential for reconstructing complex MRI sequences.