Reducing Navigators in Free-Breathing Abdominal MRI via Temporal Interpolation Using Convolutional Neural Networks

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Navigated 2-D multi-slice dynamic MRI is crucial for MR guided therapies, enabling visualization of breathing motion.
• Current methods require multiple navigator slices, limiting resolution or increasing acquisition time.
• Reducing navigator slices could enhance imaging efficiency.

Purpose of the Study:

• To develop and evaluate a convolutional neural network (CNN) for temporal interpolation of navigator slices in dynamic MRI.
• To reduce the number of acquired navigator slices while maintaining image quality.
• To improve the efficiency of navigated dynamic MRI for MR guided therapies.

Main Methods:

• Proposed temporal interpolation of navigator slices using CNNs.
• Evaluated CNN performance on abdominal navigated dynamic MRI data from 14 subjects.
• Compared CNN interpolation with non-linear registration-based interpolation.

Main Results:

• CNN-based interpolation achieved higher accuracy than non-linear registration, particularly at peak motion points.
• The CNN approach demonstrated substantially lower execution times.
• Dynamic volume reconstructions showed minimal differences between acquired and interpolated navigator slices.

Conclusions:

• CNN-based temporal interpolation is an effective method for reducing navigator slices in dynamic MRI.
• This technique can improve through-plane resolution or reduce acquisition time for MR guided therapies.
• The proposed method offers a computationally efficient and accurate alternative to registration-based interpolation.