Application of machine learning methods on acoustic physiological signals for the diagnosis and management of chronic diseases
File(s)
Author(s)
Kok, Xuen Hoong
Type
Thesis
Abstract
Objectives: Chronic diseases have massive impact on the quality of life of those who are affected, which can also extend the burden to their caregivers. The two conditions that are studied further in this thesis are chronic respiratory disease (CRD) and epilepsy. These have been selected due to the significant and rising mortality associated with the former, and the similarities in some of the most critical physiological manifestations exhibited during seizures in the latter. Early diagnosis and effective management are crucial steps toward the reduction in both disease burden and mortality rates, through long-term monitoring (LTM). Daily monitoring of patients with chronic disease, however, could only be realised efficiently in practice using non-invasive wearable or non-contact systems. Hence, in this thesis, various physiological signals, measurements, and processing methods suitable for wearable long-term data acquisition were explored. Based on these findings, the potential of acoustic signals in the context of these two diseases was demonstrated.
Methods: In this thesis, a framework for the analysis of acoustic physiological signals obtained from LTM is described. The proposed approach utilised machine learning techniques within the analysis pipeline to better leverage and recognise patterns available in long-range data. Acoustic signals were used in the two chronic diseases examined as they are direct measurements of the pulmonary function and/or obstructions in the airways of (CRD) subjects; and in epilepsy, have overlapping characteristics during seizures which allow the same sensors to be used for data acquisition at the same site.
Results and clinical significance: This thesis shows that using the proposed approach, the algorithms developed were able to successfully identify persons with CRD and chronic obstructive pulmonary disease (COPD), and in the case of epilepsy subjects, quantify and determine when seizures occur; demonstrating the feasibility of acoustic signals acquired from LTM in the diagnosis and management of chronic diseases. These findings could eventually be used to create devices that help to reduce disease burden, improving the quality of life of persons diagnosed with chronic diseases, as well as an improved disease prognosis.
Methods: In this thesis, a framework for the analysis of acoustic physiological signals obtained from LTM is described. The proposed approach utilised machine learning techniques within the analysis pipeline to better leverage and recognise patterns available in long-range data. Acoustic signals were used in the two chronic diseases examined as they are direct measurements of the pulmonary function and/or obstructions in the airways of (CRD) subjects; and in epilepsy, have overlapping characteristics during seizures which allow the same sensors to be used for data acquisition at the same site.
Results and clinical significance: This thesis shows that using the proposed approach, the algorithms developed were able to successfully identify persons with CRD and chronic obstructive pulmonary disease (COPD), and in the case of epilepsy subjects, quantify and determine when seizures occur; demonstrating the feasibility of acoustic signals acquired from LTM in the diagnosis and management of chronic diseases. These findings could eventually be used to create devices that help to reduce disease burden, improving the quality of life of persons diagnosed with chronic diseases, as well as an improved disease prognosis.
Version
Open Access
Date Issued
2022-07
Date Awarded
2022-11
Copyright Statement
Creative Commons Attribution NonCommercial Licence
License URL
Advisor
Rodriguez Villegas, Esther
Publisher Department
Electrical and Electronic Engineering
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)