Computer Assisted Cardiac Auscultation: Probabilistic Modelling and Psychoacoustic Feature Extraction for Heart Sound Descriptions

aut.embargoNoen_NZ
aut.thirdpc.containsNoen_NZ
dc.contributor.advisorLowe, Andrew
dc.contributor.advisorLegget, Malcolm
dc.contributor.authorMeintjes, Andries
dc.date.accessioned2020-05-03T23:41:13Z
dc.date.available2020-05-03T23:41:13Z
dc.date.copyright2020
dc.date.issued2020
dc.date.updated2020-05-03T09:15:36Z
dc.description.abstractPhysicians have been using stethoscopes for over 200 years to listen to the sound produced by the heart, but the diagnostic accuracy of this practice has been called into question by studies that have found the clinical skills of doctors at all levels to be lacking. The development of electronic stethoscopes and advancements in the processing power of digital computers and the subsequent development of signal processing and machine learning methods has opened the door to the field of Computer Assisted Cardiac Auscultation (CACA). This field of study is concerned with increasing the diagnostic value of the heart sounds using computers and the myriad signal processing methods that these devices enable. This thesis represents an exploration of CACA from the viewpoint of probabilistic and psychoacoustic modelling. Probability theory provides the framework with which we model the heart sounds, firstly, using an unsupervised machine learning method called Independent Component Analysis (ICA) and secondly, by expanding current work on the use of duration-dependent Hidden Markov Models (HMM). We also investigate heart sounds as perceptual phenomena using psychoacoustic models to arrive at descriptions of features of heart murmurs that correspond to those that an expert auscultator would listen for when auscultating. This enables the findings of the algorithm to be communicated in a form that is familiar and acceptable. We present four case studies on the use of ICA in which the model can identify physiologically and diagnostically interesting features in heart cycles given an appropriately chosen number of sources. A probabilistic systolic murmur labelling model is developed as an expansion of previous work done in heart sound segmentation. The proposed algorithm achieves an F1-score of 93.6% compared to 90.6% achieved by the current state-of-the-art and can identify systolic murmurs with an area under the receiver-operator curve (AUC) of 0.90 as tested on a dataset of 56 heart sound recordings. In the final part of the thesis, psychoacoustic models are developed for systolic murmurs. The perceptual qualities of 'loudness', 'pitch', and 'shape' are derived using psychoacoustic principles and compared to annotations made by expert auscultators. An online survey was developed and tested for the purpose of collecting expert annotations. The completion rate of the survey was 16%, perhaps in part due to the complex and time consuming nature of the task compounded by the online format of the survey. The collected responses show a percent agreement of 0.73 for 'loudness', 0.65 for 'pitch', and 0.35 for 'shape'. The proposed model showed strongest agreement with 'loudness' and some agreement with 'pitch', but there was little agreement on the 'shape' feature. This thesis shows that the application of ICA, the explicit modelling of a systolic murmur state in heart sound segmentation, and models of psychoacoustic features increase the diagnostic value of heart sounds.en_NZ
dc.identifier.urihttps://hdl.handle.net/10292/13306
dc.language.isoenen_NZ
dc.publisherAuckland University of Technology
dc.rights.accessrightsOpenAccess
dc.subjectBiomedical Signal Processingen_NZ
dc.subjectHeart Sound Analysisen_NZ
dc.subjectComputer Assisted Cardiac Auscultationen_NZ
dc.subjectProbabilistic Heart Sound Modelsen_NZ
dc.titleComputer Assisted Cardiac Auscultation: Probabilistic Modelling and Psychoacoustic Feature Extraction for Heart Sound Descriptionsen_NZ
dc.typeThesisen_NZ
thesis.degree.grantorAuckland University of Technology
thesis.degree.levelDoctoral Theses
thesis.degree.nameDoctor of Philosophyen_NZ
Files
Original bundle
Now showing 1 - 1 of 1
Loading...
Thumbnail Image
Name:
PhD Thesis - Andries Meintjes.pdf
Size:
42.29 MB
Format:
Adobe Portable Document Format
Description:
Thesis
License bundle
Now showing 1 - 1 of 1
Loading...
Thumbnail Image
Name:
license.txt
Size:
889 B
Format:
Item-specific license agreed upon to submission
Description:
Collections