Click here to

Session: Multi-Disciplinary General ePoster Viewing [Return to Session]

Application of Super-Resolution Reconstruction to Prostate 0.35T MRI Scans

J Charters1*, A Kishan1, J Lamb1, (1) Dept. of Radiation Oncology, UCLA, Los Angeles, CA


PO-GePV-M-185 (Sunday, 7/25/2021)   [Eastern Time (GMT-4)]

Purpose: Super-resolution (SR) is a technique to combine multiple low-resolution images acquired with sub-voxel frame shifts into a single high-resolution image by inverting a linear image acquisition model, which may be used to address resolution limitations arising from subject motion and hardware. In this study we evaluate the performance of SR on 0.35T magnetic resonance imaging (MRI) prostate scans towards an application of low-field MRI-guided focal therapy.

Methods: Balanced steady-state free precession (bSSFP) multi-slice images of an American College of Radiology (ACR) MRI accreditation phantom and one prostate patient were acquired on a 0.35T MRI-guided radiotherapy machine. All images had a pixel spacing of 1.5mm x 1.5mm. Couch translations of 0.5mm and 1.0mm were applied in both in-plane directions, for a total of five acquisitions per object. Registrations were performed using Elastix. We used a maximum a posteriori (MAP) based SR algorithm to reconstruct the reference object by a factor of 2 in each dimension. We assumed uniform Gaussian blurring in the SR forward model from high-resolution to low-resolution and a Gaussian noise prior. The resulting images were compared to bilinear upsampling and an upsampled average of the registered images. Example edges in the patient image were chosen to extract line profiles, which were then fit to a sigmoid function. The sigmoid widths quantified spatial resolution of the object.

Results: In the ACR phantom high-contrast spatial resolution slice, the SR image showed improvement in the visibility of the hole-array pairs. Similarly, greater detail was present in the prostate patient SR reconstruction. Sigmoid curve fitting demonstrated smaller widths for SR over single image and average image interpolation.

Conclusion: MAP-based SR is a feasible method to improve 0.35T MR prostate image resolution. Additional improvements, such as incorporating a measured imager point-spread function, may be beneficial.



    Not Applicable / None Entered.


    Not Applicable / None Entered.

    Contact Email