Real-time 3D motion estimation from undersampled MRI using multi-resolution neural networks

This article was originally published here

Med Phys. 2021 Sep 15. doi: 10.1002/mp.15217. Online ahead of print.


PURPOSE: To enable real-time adaptive MRI-guided radiotherapy (MRIgRT) by obtaining time-resolved 3D deformation vector fields (DVFs) with high spatio-temporal resolution and low latency (< 500 ms). Theory & Methods: Respiratory-resolved T1 -weighted 4D-MRI of 27 patients with lung cancer were acquired using a golden-angle radial stack-of-stars readout. A multi-resolution convolutional neural network called TEMPEST was trained on up to 32x retrospectively undersampled MRI of 17 patients, reconstructed with a non-uniform fast Fourier transform, to learn optical flow DVFs. TEMPEST was validated using 4D respiratory-resolved MRI, a digital phantom, and a physical motion phantom. The time-resolved motion estimation was evaluated in-vivo using two volunteer scans, acquired on a hybrid MR-scanner with integrated linear accelerator. Finally, we evaluated the model robustness on a publicly-available 4D-CT dataset.

RESULTS: TEMPEST produced accurate DVFs on respiratory-resolved MRI at twenty-fold acceleration, with the average end-point-error <2 mm, both on respiratory-sorted MRI and on a digital phantom. TEMPEST estimated accurate time-resolved DVFs on MRI of a motion phantom, with an error <2 mm at 28x undersampling. On two volunteer scans, TEMPEST accurately estimated motion compared to the self-navigation signal using 50 spokes per dynamic (366x undersampling). At this undersampling factor, DVFs were estimated within 200 ms, including MRI acquisition. On fully-sampled CT data, we achieved a target registration error of 1.87± 1.65 mm without retraining the model.

CONCLUSION: A CNN trained on undersampled MRI produced accurate 3D DVFs with high spatio-temporal resolution for MRIgRT.

PMID:34525223 | DOI:10.1002/mp.15217