Author: B. Gino Fallone, Keith D. Wachowicz, Mark G. Wright, Jihyun Yun 👨🔬
Affiliation: Medical Physics Division, Department of Oncology, University of Alberta, Dept. of Medical Physics, Cross Cancer Institute and Dept. of Oncology, University of Alberta, Medical Physics Division, Department of Oncology, University of Alberta and Department of Medical Physics, Cross Cancer Institute, Dept. of Medical Physics, Cross Cancer Institute and Dept. of Oncology, University of Alberta; MagnetTx Oncology Solutions, www.magnetTX.com 🌍
Purpose: This work focusses on a hybrid Principal Component Analysis (PCA) based MR acceleration method for prospectively acquired undersampled data on a hybrid MR-Linac system for real-time target-tracking purposes. The goal is to demonstrate positional-accuracy of a target on the reconstructed images.
Methods: Data were acquired prospectively on a hybrid MR-Linac system using an MR-compatible motion phantom at accelerated undersampling rates of 4x, 5x and 8x relative to a fully-sampled 128x128 acquisition at 2 frames/second. A sinusoidal-trajectory (amplitude 19.9mm, period 4s) was programmed and an MR-visible plastic sphere (diameter 50mm) was imaged. A pre-determined number of core phase encodes around central k-space are acquired in every frame. Outside the core, k-space is undersampled in such a way that all of k-space is sampled in a pre-determined number of frames. The hybrid descriptor of the PCA method refers to the reconstruction looking at the time-evolution of the core phase encodes along with the relationships between outer and core k-space. An important user-controlled parameter is the size of reconstruction window (Nwin) consisting of previous frames+the current frame. RMSE was used to compare the measured placement of the reconstructed phantom versus the known phantom location. Nwin values between 30 and 120 frames (15 frame-intervals at 4x/8x, 30 frame-intervals at 5x) were tested at each acceleration.
Results: RMSE values were <1mm (<0.5mm with appropriate Nwin selection) for all accelerations tested. It was found that as Nwin increased, RMSE improved. Consistent under-estimation of target location was found at lower Nwin values, however this improved as Nwin increased. At 8x acceleration, RMSE was found to be 0.99mm and 0.45mm at Nwin of 30 and 120 respectively.
Conclusion: This work demonstrates the positional-accuracy of reconstructed images from prospectively acquired data using a hybrid-PCA method. The results are promising for target-tracking purposes for use on hybrid MR-Linac systems.