Compressed Sensing for Moving Imagery in Medical Imaging

Çağdaş Bilen, Yao Wang, Ivan Selesnick

Abstract: Numerous applications in signal processing have benefited from the theory of compressed sensing which shows that it is possible to reconstruct signals sampled below the Nyquist rate when certain conditions are satisfied. One of these conditions is that there exists a known transform that represents the signal with a sufficiently small number of non-zero coefficients. However when the signal to be reconstructed is composed of moving images or volumes, it is challenging to form such regularization constraints with traditional transforms such as wavelets. In this paper, we present a motion compensating prior for such signals that is derived directly from the optical flow constraint and can utilize the motion information during compressed sensing reconstruction. Proposed regularization method can be used in a wide variety of applications involving compressed sensing and images or volumes of moving and deforming objects. It is also shown that it is possible to estimate the signal and the motion jointly or separately. Practical examples from magnetic resonance imaging has been presented to demonstrate the benefit of the proposed method.

Introduction

Compressed Sensing (CS) has revolutionized medical imaging by enabling the reconstruction of high-quality images from significantly undersampled data. This relies on the signal being sparse in a transform domain. In dynamic imaging, such as cardiac MRI, signals change over time. While standard techniques like temporal Total Variation (TV) or Fourier transforms exploit temporal redundancies, they often fail during significant motion, leading to "staircase" artifacts or blurring. Existing motion-compensated CS methods typically require high-quality reference frames, which are often unavailable when reconstructing the entire sequence simultaneously from undersampled data.

Motion-Compensated Total Variation (Motion-TV)

The proposed Motion-TV regularization incorporates motion compensation directly into the reconstruction process by leveraging the optical flow constraint. Unlike standard temporal TV, which penalizes raw differences between consecutive frames, Motion-TV penalizes the difference between a frame and the motion-compensated prediction of the previous frame.

Bilinear interpolation scheme for motion compensation between frames.

As shown in Figure 1, a pixel in frame $i$ is predicted from its corresponding location in frame $i-1$ using a motion vector $\mathbf{v}_i(s)$. Since the predicted location may not fall on a grid point, bilinear interpolation (using parameters $a$ and $b$) is used to estimate the intensity. This relationship is formalized into a motion-warping matrix $\mathbf{K}$, and the reconstruction algorithm seeks a signal that minimizes the sparsity of this motion-compensated residual.

Optimization can be performed separately—by first estimating motion from a low-quality reconstruction—or jointly, where image and motion vectors are updated iteratively. The Alternating Direction Method of Multipliers (ADMM) is employed to solve these complex optimization problems efficiently.

Experimental Results

The method was evaluated on two 2D cardiac MRI cine datasets. Dataset 1 consisted of 22 frames (128x128), while Dataset 2 had 24 frames (256x256) and exhibited larger motion. Both were retrospectively undersampled at rates up to R=14.

RMSE in ROI for each frame of Dataset 2, comparing various reconstruction methods at R=8 and R=14.

Quantitative analysis (Figure 2) demonstrates that Motion-TV and Joint Motion-TV consistently achieve lower Root Mean Squared Error (RMSE) compared to standard temporal DFT and TV methods, especially at high acceleration factors. In Dataset 2, where motion is more pronounced, the performance gap is even more significant. Qualitatively, Motion-TV preserves sharper edges of the heart wall and avoids the artifacts typical of non-motion-compensated methods.

Conclusion and Future Work

Motion-TV provides a robust framework for dynamic CS reconstruction by integrating motion information directly into the prior. While it requires hyperparameter tuning and currently uses bilinear interpolation, it shows significant promise for high-acceleration cardiac imaging. Future work includes exploring higher-order interpolation and applying the method to correct for respiratory or bulk patient motion in various clinical scenarios.

Summary updated

Abstract

Numerous applications in signal processing have benefited from the theory of compressed sensing which shows that it is possible to reconstruct signals sampled below the Nyquist rate when certain conditions are satisfied. One of these conditions is that there exists a known transform that represents the signal with a sufficiently small number of non-zero coefficients. However when the signal to be reconstructed is composed of moving images or volumes, it is challenging to form such regularization constraints with traditional transforms such as wavelets. In this paper, we present a motion compensating prior for such signals that is derived directly from the optical flow constraint and can utilize the motion information during compressed sensing reconstruction. Proposed regularization method can be used in a wide variety of applications involving compressed sensing and images or volumes of moving and deforming objects. It is also shown that it is possible to estimate the signal and the motion jointly or separately. Practical examples from magnetic resonance imaging has been presented to demonstrate the benefit of the proposed method.
BibTeX
@misc{bilen2012compressed,
  author = {Çağdaş Bilen and Yao Wang and Ivan Selesnick},
  title = {{Compressed Sensing for Moving Imagery in Medical Imaging}},
  year = {2012},
  eprint = {1203.5772},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1203.5772},
}
Figure