Click here to

Session: [Return to Session]

Koopman Theory-Based Dynamic Mode Decomposition for Dynamic Contrast-Enhanced HNC MRI Motion Correction

R He, K Wahid, A Mohamed*, B McDonald, Y Ding, M Naser, J Wang, K Hutcheson, C Fuller, S Lai, MD Anderson Cancer Center, Houston, TX

Presentations

PO-GePV-M-247 (Sunday, 7/10/2022)   [Eastern Time (GMT-4)]

ePoster Forums

Purpose: Dynamic contrast-enhanced (DCE) MRI plays an import role in head and neck cancer (HNC) studies but it plagued by errors introduced by motion. To suppress the motion in HNC DCE MRI, we used Koopman theory based dynamic mode decomposition (DMD) to separate motion from DCE signal.

Methods: In HNC DCE MRI, voxels in regions affected by motion show sudden intensity deviation in their time-course curves and spatial deformation in shape. This not only disrupts DCE data fitting, but also makes the determination of arterial input function (AIF) difficult. Together they causes additional uncertainties in the estimation procedures. The common method of deformable image registration is not effective in HNC, aside from the burden in the computation of volumetric time series, the image in motion is usually blurry and noisy. DMD analyzes the dynamic systems without any prior domain knowledge but only the data using a linear approximation originated from Koopman analysis. The Koopman operator is an infinite linear operator that lifts the state of the dynamic systems to an observable space where the original nonlinear dynamic becomes linear; it can work on finite space by approximation. DMD can find the hidden structure of the linear system and decompose the state to a series of modes associated with a spatial-temporal pattern and eigenvalue indicating the evolution of the pattern. That provide a better motion suppression approach for HNC DCE MRI.

Results: We practice DMD and Koopman analysis on clinical HNC DCE MRI and prove the motion can be separated from images. However, we used the hand crafted Koopman operator adapted from Extended DMD that the intensity of DCE is not fully recovered.

Conclusion: DMD and Koopman analysis is promising for motion suppression for clinical HNC DCE MRI, using Deep Learning to learn the Koopman approximation will improve the performance.

Keywords

Not Applicable / None Entered.

Taxonomy

IM/TH- Image Registration Techniques: Modality: MRI

Contact Email

Share: