The Library
Deep learning for electrocardiogram analysis applied to health monitoring applications
Tools
Porumb, Mihaela (2020) Deep learning for electrocardiogram analysis applied to health monitoring applications. PhD thesis, University of Warwick.
|
PDF
WRAP_Theses_Porumb_2020.pdf - Submitted Version - Requires a PDF viewer. Download (53Mb) | Preview |
Official URL: http://webcat.warwick.ac.uk/record=b3494931~S15
Abstract
Several recent advances fuelled the significant increase in interest for Artificial Intelligence-based healthcare innovations: the vast availability of biomedical data, accurate wearable sensors, electronic health records, the advancement in Machine Learning methods and affordable computational resources. This thesis focuses on electrocardiogram signal analysis using a range of deep learning techniques, including Convolutional Neural Networks, Recurrent Neural Networks and Convolutional Autoencoders for developing several health monitoring applications. The performance of the proposed models was investigated in two applications (i.e. nocturnal low glucose detection and congestive heart failure diagnosis), that share the same aspects of the input data including noise, time-dependence, heterogeneity. This thesis explores the efficacy of a personalised deep learning-based system for the automatic detection of low glucose levels in real-life settings, overcoming the limitations of previous attempts. Furthermore, this thesis explores unsupervised methods for learning time series representations to address the high cost and scarcity of the labelled biomedical data. A novel deep learning-based method that employed raw electrocardiogram signals was explored for congestive heart failure diagnosis. Added to their superior performance, the results presented in this thesis bring forward the intuition that short electrocardiogram recordings, of just about 5 minutes, can be sufficient to diagnose correctly severe congestive heart failure. A third case study presented in this thesis advances the idea that certain heart rate variability features can be estimated from a short (≈ 1 minute) raw electrocardiogram signal, which might facilitate the development of real-time applications.
This thesis shows that leveraging deep learning-based models for physiological signal analysis, not only bypasses the costly feature extraction and selection process, but they can reveal unintuitive patterns in the input data that are class-discriminative. Providing transparency on how the models reached certain conclusions is crucial for the healthcare field, firstly because the medical professionals expect specific evidence for their recommendations and secondly, the models can help doctors better understand the physiological processes.
Overall, the findings of this thesis confirmed that deep learning methods applied for electrocardiogram analysis could improve and extend current diagnostic models, might bring new research opportunities and contribute to the development of real-time health monitoring applications.
Item Type: | Thesis (PhD) | ||||
---|---|---|---|---|---|
Subjects: | Q Science > Q Science (General) R Medicine > R Medicine (General) R Medicine > RC Internal medicine |
||||
Library of Congress Subject Headings (LCSH): | Electrocardiography -- Data processing, Machine learning -- Mathematical models, Congestive heart failure -- Diagnosis, Hypoglycemia -- Diagnosis, Patient monitoring -- Data processing | ||||
Official Date: | February 2020 | ||||
Dates: |
|
||||
Institution: | University of Warwick | ||||
Theses Department: | School of Engineering | ||||
Thesis Type: | PhD | ||||
Publication Status: | Unpublished | ||||
Supervisor(s)/Advisor: | Pecchia, Leandro ; Hattersley, John G. | ||||
Format of File: | |||||
Extent: | xxi, 224 leaves : illustrations (chiefly colour) | ||||
Language: | eng |
Request changes or add full text files to a record
Repository staff actions (login required)
View Item |
Downloads
Downloads per month over past year