Deblurring for Spiral Real-Time MRI Using Convolutional Neural Networks

Purpose: To develop and evaluate a fast and effective method for deblurring spiral realtime magnetic resonance imaging (RT-MRI) using convolutional neural networks (CNN). Methods: We demonstrate a 3-layer residual CNN to correct image domain off-resonance artifacts in speech production spiral RT-MRI without knowledge of field maps. The architecture is motivated by the multi-frequency segmentation method. Spatially varying off-resonance blur is synthetically generated by using discrete object approximation and field maps with data augmentation from a large database of 2D human speech production RT-MRI. The impact of off-resonance frequency range, appearance of blurring, and readout durations is validated with synthetic data. The proposed method is tested using synthetic test and in vivo data with longer readouts, quantitatively using image quality metrics and qualitatively via visual inspection, and via comparison with conventional deblurring methods. Results: Deblurring performance was found superior to a current auto-calibrated method for in vivo data, and only slightly worse than ideal reconstruction with perfect knowledge of the field map for synthetic test data. CNN deblurring made it possible to visualize articulator boundaries with readouts up to 8 ms at 1.5 T, which is 3-fold longer than the current standard practice. The computation time was 12.3 + 2.2 ms per-frame, enabling a low-latency processing for RT-MRI applications. Conclusion: CNN-deblurring is a practical, efficient, and field-map-free approach for deblurring of spiral RT-MRI. In the context of speech production imaging this can enable 1.7-fold improvement in scan efficiency and the use of spiral readouts at higher field strengths such as 3 T.