Speaker: Yeun Kim
Abstract
Three-dimensional ultrasound speckle tracking enables volumetric estimation of tissue motion, but its practical use is limited by the high computational cost of matching local speckle patterns in 3D data. Compared with 2D tracking, volumetric tracking requires larger search spaces and greater memory usage, making real-time performance difficult to achieve. This thesis investigates FFT-based correlation and normalized cross-correlation for GPU-accelerated 3D ultrasound speckle tracking. The proposed approach estimates displacement by comparing local reference blocks with candidate blocks in subsequent ultrasound volumes. FFT-based computation is used to accelerate correlation operations, while GPU implementation addresses the computational demands of volumetric data. The implementation is evaluated in terms of runtime, scalability, memory usage, and motion-estimation performance. The thesis aims to identify the practical trade-offs of NCC-based 3D speckle tracking.