Computer Vision Elastography
Algorithm Implementation Outline
This work provides a comprehensive study of a hierarchical region-based motion estimation technique, using novel refinements to resolve existing speckle tracking issues. The basis of our approach uses multiple hierarchical region-based matching, encapsulating more unique speckle pattern without restricting spatial resolution. From our earlier experiments, multiscale template analysis produced consistently superior results than single scale or biased shaped and sized templates that correspond to the shape of the object to be tracked. We extend our previous interframe displacement work by using trajectories, which quantify continuous temporal displacement of speckle movement for complete sequences. This improves upon the prominent approach of specifically analysing manually selected frame pairs, as trajectories incorporate interframe displacements for every frame pair for all sequences, yielding measurements of local temporal movement. Trajectory drift correction is also applied, to ensure regions are accurately tracked in space and time. The proposed method corrects for potential freehand transducer motion, and updates for possible changes in the image plane as a result of axial, lateral and elevation movement. To improve displacement accuracy, displacements are refined by automatically selecting between two matching measures to compensate for speckle noise. The first measure uses the normalised cross correlation (NCC) for regions of strong signal (refined to minimise aperture problems), and the second measure a speckle formulated similarity measure is applied in regions of low SNR from speckle, determined by the SNR. Finally we determine displacement field accuracy, and demonstrate trajectory homogeneity by exploiting the trajectory path coherence, using the output as a validation measure.
Several strategies are developed to improve on the estimated displacements produced from block matching methods presented in the previous works. These include:
- To ensure the appropriate use of similarity measures in regions of varying signal content, a novel alternating similarity measure strategy is introduced, which applies either the normalised cross correlation in regions of tendon (low speckle), or the speckle measure in regions of soft tissue (dense speckle), indicated by the local image SNR.
- To improve on specific frame pair analysis, interframe displacements are extended to trajectories that track regions of signal through a sequence, utilising strategies for:
- Trajectory drift correction, to eliminate temporal displacement error from propagating displacement inaccuracy.
- Trajectory updating, to adjust the trajectory fields to include movement in the 2D image caused from movement in the 3D volume.
- Trajectory path coherence, to provide a means of measuring and comparing a trajectory fields temporal displacement activity.