Towards Automatic and Consistent Cardiac Analysis
31 Jan 2018 Wednesday, 10:00 AM to 11:30 AM
COM2 Level 4
Executive Classroom, COM2-04-02
Analysis of the cardiac function and identification of any potential cardiovascular disease is often performed using imaging technologies such as magnetic resonance imaging (MRI). After segmenting the heart from the acquired images, evaluation and analysis is carried out by certain clinical parameters and regional indicators that are calculated from the extracted heart. Diagnosis is highly sensitive to the segmentations and other image-driven inputs. Consequently, significant steps have been taken towards providing effective tools for automatic analysis of the heart with the goal of improving accuracy of the evaluations and reproducibility of the results as well as decreasing the processing time. There remains, however, many steps in the evaluation process that still lack automation or accuracy. Examples of such cases are selection of the image slices which are within the left ventricle or accurate myocardium strain analysis in a particular imaging technique called myocardium tagging.
This thesis makes three contributions to automating and improving different steps of the cardiac analysis procedure. We present a novel approach for identifying the most basal image slice within the left ventricle using the long-axis view of the heart. This is achieved by using a model-based segmentation approach that trains an active shape model for segmentation of the long-axis view of the left ventricle and identifies the slices of interest based on these segmentations. Our second contribution is to develop an image-based segmentation approach for selection of the slices of interest which eliminates the need for training.
A reliable myocardium strain analysis requires both accurate motion tracking and correct annotation of the myocardium during the cardiac cycle. While many algorithms have been proposed for accurate tracking of the heart in tagged MR images, little focus has been placed on ensuring correct annotation of the tagged myocardium during the cardiac cycle. As the third contribution, we propose a method to improve reliability of myocardium strain analysis in tagged MR images by leveraging the available high-quality cine MRI sequence. The cine MRI segmentation is utilized to generate a series of myocardium tracking proposals from which the myocardium tracking that annotates the myocardium boundaries accurately during the cardiac cycle is selected and used for myocardium strain analysis.