Click here to

Session: Multi-Disciplinary: MRI in Radiation Therapy [Return to Session]

Improving DWI Spatial Resolution with Respiratory Motion Modeling On An Interleaved BSSFP-DWI Sequence

J Charters1*, D O'Connell1, Y Gao1, Y Yang1, J Ginn2, J Lamb1, (1) University of California, Los Angeles, Los Angeles, CA (2) Washington University, Saint Louis, MO


MO-IePD-TRACK 3-3 (Monday, 7/26/2021) 5:30 PM - 6:00 PM [Eastern Time (GMT-4)]

Purpose: Diffusion-weighted imaging (DWI) may serve as a biomarker for lung and pancreatic tumors undergoing magnetic resonance imaging (MRI)-guided radiotherapy (MRgRT). However, obtaining DWI with sufficient signal-to-noise ratio (SNR) and resolution for biomarker use is challenging for these tumors due to substantial respiratory motion. Furthermore, direct co-registration and summation of DWI frames of moving tumors is difficult due to lack of anatomic information. Here we demonstrate a motion deblurring technique based on interleaving high-resolution, high SNR balanced steady-state free precession (bSSFP) images with low-resolution, low SNR DWI images.

Methods: Studies on a dynamic thorax phantom and one volunteer were performed on a 0.35T MRgRT machine. The subjects experienced a sequence of bSSFP and DWI multislice scans, which repeated 100 times over approximately 17 minutes in order to adequately capture respiratory motion. A respiratory bellows surrogate was processed in LabVIEW and served as the motion surrogate. The bSSFP images and respiratory bellows were used to extract a motion model which related tissue motion to the bellows trace. The model coefficients were resampled from the bSSFP grid (2mm x 2mm spacing) to the DWI grid (2.73mm x 2.73mm spacing). The warped coefficients, together with bellows surrogate values corresponding to the DWI images, were used as estimated DWI deformation vector fields (DVF). Signal averaging across all DWI images was compared to averaging across the model-registered DWI images. Resolution enhancements were evaluated visually and quantified by fitting a sigmoid function to line profiles across intensity edges.

Results: Model-estimated DVF accuracy on the bSSFP phantom images was subpixel for nearly every image. Line profile sigmoid widths for the phantom were 8.3 and 3.6 pixels for non-registered and phantom-registered images. Visually, motion blurring was reduced.

Conclusion: Our work demonstrates the feasibility of motion de-blurring of DWI images using a motion model built from interleaved bSSFP images.



    Not Applicable / None Entered.


    Not Applicable / None Entered.

    Contact Email