Myocardial contrast echocardiography (MCE) uses microbubble contrast agents and ultrasound
to visualise myocardial vasculature and perfusion in the human body. MCE perfusion analysis
can aid in the diagnosis of coronary artery disease (CAD). However, most analyses rely on
human visual assessments which are subjective and operator-dependent. There is a strong
need to develop automatic MCE quantification methods but this is hindered by 1) the high
variabilities in MCE data and 2) the lack of robust and easy-to-use computerised quantification
tools. The thesis aims to develop automatic segmentation and quantification methods for fast,
accurate and operator-independent MCE perfusion assessment in order to aid the diagnosis of
CAD.
Firstly, a fully automatic approach is developed for fast and accurate myocardial segmentation
in 2D MCE sequence. The method is primarily based on the random forest framework
and is additionally constrained by a statistical shape model of the myocardium to improve
the final segmentation. When evaluated on human MCE sequences, our proposed approach
produces accurate segmentation results, outperforming the other state-of-the-art segmentation approaches.
Secondly, a robust and easy-to-use software is developed specifically for MCE perfusion quantification. The software allows for semi-automatic myocardial segmentation, attenuation correction, automatic rejection of poor quality data and automatic perfusion quantification. The
developed software demonstrates good CAD diagnostic performance comparable to that of
MCE visual assessments and SPECT. Perfusion quantification using the software is also fast,
reproducible and less operator-dependent.
Finally, two advanced MCE quantification techniques are explored. The first involves developing more sophisticated perfusion models for MCE quantification based on knowledge of physiology and ultrasound physics. The second involves employing a support vector machine for
CAD detection based on the perfusion parameters extracted from the various perfusion models.
The combination of these two techniques has improved the accuracy of CAD diagnosis.