An improved classification approach for echocardiograms embedding temporal information

Li, Wei (2016) An improved classification approach for echocardiograms embedding temporal information. PhD thesis, Middlesex University. [Thesis]

PDF - Final accepted version (with author's formatting)
Download (5MB) | Preview


Cardiovascular disease is an umbrella term for all diseases of the heart. At present, computer-aided echocardiogram diagnosis is becoming increasingly beneficial. For echocardiography, different cardiac views can be acquired depending on the location and angulations of the ultrasound transducer. Hence, the automatic echocardiogram view classification is the first step for echocardiogram diagnosis, especially for computer-aided system and even for automatic diagnosis in the future. In addition, heart views classification makes it possible to label images especially for large-scale echo videos, provide a facility for database management and collection.

This thesis presents a framework for automatic cardiac viewpoints classification of echocardiogram video data. In this research, we aim to overcome the challenges facing this investigation while analyzing, recognizing and classifying echocardiogram videos from 3D (2D spatial and 1D temporal) space. Specifically, we extend 2D KAZE approach into 3D space for feature detection and propose a histogram of acceleration as feature descriptor. Subsequently, feature encoding follows before the application of SVM to classify echo videos.

In addition, comparison with the state of the art methodologies also takes place, including 2D SIFT, 3D SIFT, and optical flow technique to extract temporal information sustained in the video images.

As a result, the performance of 2D KAZE, 2D KAZE with Optical Flow, 3D KAZE, Optical Flow, 2D SIFT and 3D SIFT delivers accuracy rate of 89.4%, 84.3%, 87.9%, 79.4%, 83.8% and 73.8% respectively for the eight view classes of echo videos.

Item Type: Thesis (PhD)
Research Areas: A. > School of Science and Technology
A. > School of Science and Technology > Computer Science
B. > Theses
Item ID: 21260
Depositing User: Jennifer Basford
Date Deposited: 09 Feb 2017 16:24
Last Modified: 21 Jun 2021 16:47

Actions (login required)

View Item View Item


Activity Overview

Additional statistics are available via IRStats2.