|M.Sc Student||Nataf Rebecca|
|Subject||Decomposing Echocardiographic Cines into Anatomical|
Visual Representation and Functional Data
|Department||Department of Biomedical Engineering||Supervisors||Professor Emeritus Dan Adam|
|Dr. Zvi Friedman|
In cardiology, ultrasound is usually the preferred imaging modality. Echocardiography sequences are used by the clinicians both for visual observation of the heart anatomy and for functional analysis. However, ultrasound images are strongly affected by speckle noise, caused by interferences between ultrasound waves. This is not the case in magnetic resonance imaging where images are smoother, edges are sharper and continuous, and contrast is higher. Nevertheless, the speckles in echocardiographic sequences enable automatic tracking of myocardial deformations that allow estimation of regional strains by (semi)automatic algorithms.
In this work, the objectives were to study the feasibility of 1) depicting the anatomical information present in echocardiography sequences in a form similar to that produced by MRI, and 2) representing the data necessary for strain measurement in a compressed form.
A signal processing algorithm was developed to provide such a decomposition. The algorithm was mainly based on an original use of autoregressive modeling of the ultrasound spectrum, which provided a smoothed representation of the myocardial walls while conserving edges. Using this new representation together with the original image, only the outlier reflectors (i.e. the values that protrude from the anatomical background) were then extracted. This smaller amount of data constituted the input for the semi-automatic strain measurement algorithms.
The algorithm was tested on 56 echocardiographic sequences of a full heartbeat, acquired from 29 patients in apical views. The two outputs (i.e. the smoothed anatomical representation and the “outliers-only” representation) were assessed with respect to their assigned new roles. The quality of the anatomical representation was assessed using semi-automatic segmentation algorithms that were applied to the processed data. In addition, the ability of measuring the ventricular end diastolic and end systolic volumes from these images was assessed by comparing the results obtained from 35 echocardiographic sequences to the results obtained from their corresponding original sequences. The results demonstrate a high correlation (around 0.9) between the two measurements. The assessment of the second output (i.e. “outliers-only”) was performed by a comparison between functional analysis of the “outliers-only” sequences and of the original data. This assessment provided excellent results: the process conserves the tracking in 93% of the cases, and allows measuring the global longitudinal systolic peak strain (a very important clinical measure) with a very low (0.15%) bias.
Thus, the decomposition of echocardiographic sequences, as performed in this study, was shown to be a feasible technique, providing useful anatomical representation and compressed functional data, which may support clinical diagnosis.