Computer Vision ElastographyJames Revell, Computer Vision Elastography. PhD thesis. University of Bristol, Department of Computer Science. April 2005. PDF, 15308 Kbytes.
This thesis is concerned with developing a two-dimensional (2D) ultrasound speckle tracking technique to quantify 2D axial and lateral strain fields for studying tissue dynamics. Knowledge of tissue displacement to infer strain characteristics is of major clinical importance. In part, this is due to the lack of simple and accurate non-invasive techniques to measure in vivo strain. The established technique for evaluating tissue behaviour using ultrasound is elastography, with conventional methods analysing the raw radio-frequency (RF) data to measure 1D displacements. In contrast, this work develops and applies image processing techniques to analyse the ultrasound images to measure 2D displacements. Measuring meaningful displacement descriptors using ultrasound assumes that speckle motion replicates small tissue deformation.
Current research using ultrasound image processing frequently concentrates on matching selected regions from frame pairs, an extension of RF elastography, to solve this correspondence problem. To assess the feasibility and effectiveness of displacement estimation using ultrasound images, a variable-sized block matching algorithm with a hierarchical full search is proposed. Unlike existing research, development and experimentation focuses on dynamic high frequency ultrasound of the musculoskeletal system. The aim is to provide a non-invasive in vivo approach to measuring displacement and strain in tendons and soft tissue for assessing tendinopathy.
With an emphasis on improving block matching approaches, enhancements are explored to extend this framework to sequences, generating trajectories. Further, novel contributions are incorporated to solve known speckle tracking problems. An automatic similarity measure selection is developed, adapting to varying speckle density and strain induced speckle decorrelation. Trajectories are also updated for moving objects in the scanned 3D volume that traverse the 2D plane. Other important challenges dealt with in this work include eliminating user frame pair pre-selection, freehand scanning registration, minimising tracking drift error and measuring displacement accuracy.
Finally, displacement fields are used to derive corresponding axial, lateral and shear strain elastograms. Further, the trajectory fields are used to generate spatiotemporal elastograms, producing object strain histories. The latter enables clinicians to have a unique means of comparing the knowledge of the applied motion to the strain that occurred during the activity, potentially aiding diagnosis. In addition, all results are validated using purpose-made tissue mimicking phantoms and in vitro tendon data, prior to analysing in vivo clinical sequences.