Abstract
Rigid slice-to-volume registration is a challenging task, which finds application in medical imaging problems like image fusion for image guided surgeries and motion correction for volume reconstruction. It is usually formulated as an optimization problem and solved using standard continuous methods. In this paper, we discuss how this task be formulated as a discrete labeling problem on a graph. Inspired by previous works on discrete estimation of linear transformations using Markov Random Fields (MRFs), we model it using a pairwise MRF, where the nodes are associated to the rigid parameters, and the edges encode the relation between the variables. We compare the performance of the proposed method to a continuous formulation optimized using simplex, and we discuss how it can be used to further improve the accuracy of our approach. Promising results are obtained using a monomodal dataset composed of magnetic resonance images (MRI) of a beating heart.
Introduction
Slice-to-volume registration is the process of finding the exact position and orientation of a 2D tomographic slice (such as an ultrasound or specific MRI plane) within a 3D anatomical volume. This is critical for image-guided surgery, where real-time 2D intra-operative images are fused with high-resolution pre-operative 3D scans, and for motion correction during 3D image acquisition. Standard continuous optimization methods, such as gradient descent or the simplex method, are prone to getting trapped in local minima if the initial alignment guess is poor. This paper proposes a robust alternative using discrete optimization.
Methodology
The authors propose a graph-based discrete formulation of the registration problem. The six parameters of rigid transformation (three for rotation, three for translation) are modeled as nodes in a pairwise Markov Random Field (MRF). While the relationship between these parameters is complex, a pairwise approximation is used to maintain mathematical tractability.
Instead of searching a continuous space, the algorithm samples a set of discrete variations (labels) around the current estimate. The FastPD algorithm is used to find optimal labels that minimize dissimilarity. To improve accuracy and range, a Gaussian pyramid and an incremental search strategy are employed. Finally, a Simplex Refinement step is added, using the discrete result as a high-quality starting point for a standard continuous optimizer to achieve sub-pixel precision.
Experimental Results
The method was evaluated using a monomodal dataset of 20 Cardiac MRI sequences ($192 \times 192 \times 11$ voxels). 100 random 2D slices were generated to serve as ground-truth tests.
Performance Comparison
Three approaches were compared: Simplex (purely continuous), Discrete (purely MRF), and Refined (MRF + Simplex). The Refined method outperformed the others across all metrics, achieving a Mean Absolute Difference (MAD) in intensity of 36.45, compared to 51.88 for the standard Simplex approach.
Robustness and Capture Range
In tests with poor initializations, the Simplex method often failed completely, while the Discrete and Refined methods consistently found the correct alignment. In temporal series tests simulating surgery, the discrete methods maintained a low error rate over time, whereas the Simplex method's error drifted significantly as the heart's motion increased.
Conclusion and Future Work
The hybrid discrete-continuous approach offers a robust solution for rigid slice-to-volume registration, significantly improving the "capture range" over traditional methods. However, it currently only handles rigid transformations and has a higher computational cost, taking approximately 9 minutes per registration compared to 30 seconds for the continuous version. Future work will focus on extending the framework to include linear transformations (scaling and shearing) and optimizing computational speed.
Summary updated