Automated Echocardiographic Image Interpretation Using Artificial Intelligence

Azarmehr, Neda (2021) Automated Echocardiographic Image Interpretation Using Artificial Intelligence. PhD thesis, University of Lincoln.

Documents
Automated Echocardiographic Image Interpretation Using Artificial Intelligence
Thesis
[img]
[Download]
Thesis submission form
Thesis submission form
[img]
[Download]
[img]
Preview
PDF
AzarmehrNeda_Computer Science_September 2021.pdf - Whole Document

9MB
[img]
Preview
PDF
Research Electronic Thesis Submission Form_NA (1).pdf - Supplemental Material

156kB
Item Type:Thesis (PhD)
Item Status:Live Archive

Abstract

In addition to remaining as one of the leading causes of global mortality, cardio vascular disease has a significant impact on overall health, well-being, and life expectancy. Therefore, early detection of anomalies in cardiac function has become essential for early treatment, and therefore reduction in mortalities. Echocardiography is the most commonly used modality for evaluating the structure and function of the heart. Analysis of echocardiographic images has an important role in the clinical practice in assessing the cardiac morphology and function and thereby reaching a diagnosis. The process of interpretation of echocardiographic images is considered challenging for several reasons. The manual annotation is still a daily work in the clinical routine due to the lack of reliable automatic interpretation methods. This can lead to time-consuming tasks that are prone to intra- and inter-observer variability. Echocardiographic images inherently suffer from a high level of noise and poor qualities. Therefore, although several studies have attempted automating the process, this re-mains a challenging task, and improving the accuracy of automatic echocardiography interpretation is an ongoing field. Advances in Artificial Intelligence and Deep Learning can help to construct an auto-mated, scalable pipeline for echocardiographic image interpretation steps, includingview classification, phase-detection, image segmentation with a focus on border detection, quantification of structure, and measurement of the clinical markers. This thesis aims to develop optimised automated methods for the three individual steps forming part of an echocardiographic exam, namely view classification, left ventricle segmentation, quantification, and measurement of left ventricle structure. Various Neural Architecture Search methods were employed to design efficient neural network architectures for the above tasks. Finally, an optimisation-based speckle tracking echocardiography algorithm was proposed to estimate the myocardial tissue velocities and cardiac deformation. The algorithm was adopted to measure cardiac strain which is used for detecting myocardial ischaemia. All proposed techniques were compared with the existing state-of-the-art methods. To this end, publicly available patients datasets, as well as two private datasets provided by the clinical partners to this project, were used for developments and comprehensive performance evaluations of the proposed techniques. Results demonstrated the feasibility of using automated tools for reliable echocardiographic image interpretations, which can be used as assistive tools to clinicians in obtaining clinical measurements.

Keywords:Medical imaging technology, artificial intelligence, cardiac imaging
Subjects:G Mathematical and Computer Sciences > G700 Artificial Intelligence
G Mathematical and Computer Sciences > G400 Computer Science
Divisions:College of Science > School of Computer Science
ID Code:49493
Deposited On:23 May 2022 09:43

Repository Staff Only: item control page