Motion Modelling for Respiratory Motion Estimation in Minimally Invasive Cardiac Interventions Using Intraprocedure Ultrasound Data

Abstract

Respiratory motion of the heart limits the utility of image-guided cardiac interventions, causing misalignments between the pre-procedure information used for guidance and the intra-procedure moving anatomy and instruments. As a result, the guidance can be misleading, compromising the accuracy and success of the intervention. Respiratory motion models have been proposed to estimate and correct for respiratory motion, but to date their clinical uptake has been very limited due to a lack of accuracy and robustness, and the interruptions that they typically introduce into the clinical work flow. The scope of this project was to devise methods to address these limitations and foster the clinical translation of respiratory motion models.

A novel Bayesian respiratory motion model was developed in the first part of the project. The Bayesian framework enables the combination of the robustness of a pre-procedure motion model derived from Magnetic Resonance Imaging with the intraprocedure information provided by 3D echography (echo) images. The main novelties of the approach lie in its probabilistic formulation and its ability to adapt to variable breathing patterns. The Bayesian motion model was further evaluated using live 2D echo images, proving to be accurate using both 2D and 3D echo images. Furthermore, a new motion model-driven echo acquisition framework was developed to acquire 2D echo images that automatically compensates for respiratory motion.

The second part of the project addressed the limitations associated with the dynamic calibration scan used to derive the motion model, the acquisition of which causes interruptions to the clinical work flow. A personalisation framework for population based motion models that uses anatomical features to predict cardiac respiratory motion was developed. Results show an average value for the 50th and 95th quantiles of the estimation error of 1:6mm and 4:7mm respectively, without the need for a subject-specific dynamic calibration scan.

Finally, the above mentioned parts were combined to produce a personalised Bayesian motion model. The technique is accurate and does not significantly complicate the clinical workflow, thus making it suitable for clinical uptake.

Date of Award	2014
Original language	English
Awarding Institution	King's College London
Supervisor	Andrew King (Supervisor) & Graeme Penney (Supervisor)