Nonlinear phonocardiographic Signal Processing

The stethoscope is a recognized icon for the medical profession, and for a long time, physicians have relied on auscultation for detection and characterization of cardiac disease. New advances in cardiac imaging have however changed this picture. Echocardiography and magnetic resonance imaging (MRI) have become so dominating in cardiac assessment that the main use of cardiac auscultation is nowadays as a preliminary test in the primary health care.

The aim of this thesis work has been to develop signal analysis methods for a computerized cardiac auscultation system, the intelligent stethoscope. In particular, the work focuses on classification and interpretation of features derived from the phonocardiographic (PCG) signal by using advanced signal processing techniques.


1 Introduction
1.1 Preliminaries on cardiac sounds
1.2 Preliminaries on PCG signal processing
1.3 Data sets
1.4 Outline of the thesis
1.5 Contributions
2 Origin of Heart Sounds and Murmurs
2.1 Cardiovascular anatomy and physiology
2.1.1 The heart valves
2.1.2 The cardiac electrical system
2.1.3 The cardiac cycle and the pressure-volume loop
2.1.4 Coupling in the cardiovascular system
2.1.5 Fractal physiology
2.2 Valvular heart diseases
2.3 Auscultation and phonocardiography
2.3.1 Terminology for describing cardiac sounds
2.3.2 Phonocardiography (PCG)
2.4 Acquisition of PCG signals
2.5 Flow-induced sound and vibrations
2.5.1 Heart sounds
2.5.2 Murmurs and bruits
xi2.6 Models of cardiac sound
2.6.1 Modeling the first heart sound
2.6.2 Modeling the second heart sound
2.6.3 Animal models and veterinary applications
3 Signal Processing Framework
3.1 Linear correlations and the power spectrum
3.2 Higher order statistics
3.3 Waveform complexity analysis
3.3.1 Waveform fractal dimension
3.3.2 Spectral slope
3.3.3 Entropy
3.4 Reconstructed state space analysis
3.4.1 Characterizing reconstructed state spaces
3.4.2 Dimension analysis
3.4.3 Lyapunov exponents
3.4.4 Entropy
3.5 Neural networks
3.6 Analysis of nonstationary signals
3.6.1 Joint time-frequency representations
3.6.2 Nonlinear and nonstationary signal analysis
3.7 Noise reduction
3.7.1 Ensemble averaging
3.7.2 Wavelet denoising
3.7.3 State space based denoising
3.8 Prediction
3.9 Classification
3.10 Feature selection
3.10.1 Feature ranking
3.10.2 Feature subset selection
3.11 System evaluation
3.11.1 Estimating classifier accuracy
4 Heart Sound Localization and Segmentation
4.1 Properties of heart sounds
xii4.2 Indirect heart sound localization and segmentation
4.2.1 Accurate localization of S1
4.3 Direct heart sound localization
4.3.1 Algorithm components
4.3.2 Evaluation data
4.3.3 Determination of design parameters
4.3.4 Frequencies and wavelets
4.3.5 Quadratic measures
4.3.6 Complexity based measures
4.3.7 Multi-feature heart sound localization
4.3.8 Comparison between methods
4.4 Heart sound classification
4.5 Finding the third heart sound
5 Assessing and Classifying Systolic Murmurs
5.1 Assessing and classifying systolic ejection murmurs
5.1.1 Pre-processing
5.1.2 Frequency based features
5.1.3 Nonlinear features
5.1.4 Classifying AS from physiological murmurs
5.1.5 Additional comments
5.2 Assessing and classifying regurgitant systolic murmurs
5.2.1 Pre-processing
5.2.2 Features
5.2.3 MI assessment
5.2.4 Distinguishing severe MI
5.2.5 Additional comments
5.3 Classifying murmurs of different origin
5.3.1 Features
5.3.2 Feature selection
5.3.3 Classification
6 Heart Sound Cancellation from Lung Sound Recordings
6.1 Heart sound localization
6.2 Prediction
7 Cardiovascular Time Intervals
7.1 Continuous monitoring of blood pressure changes
7.1.1 Extraction of transit times
7.1.2 Agreement between transit times and blood pressure
7.2 Respiration monitoring
7.2.1 Agreement between transit times and respiration
7.3 Additional comments
8 Complementary Remarks and Future Aspects
8.1 Areas of application
8.2 Limitations
8.2.1 Clinical validation
8.2.2 Computational complexity
8.2.3 Stationarity
8.2.4 Chaos or noise?
8.3 Future work
8.3.1 Creating a murmur map
8.3.2 Feature extraction, classification and beyond
8.3.3 The forest and the trees
8.3.4 Information fusion
8.3.5 Model-based signal analysis
8.3.6 Obstacles
8.4 Starting all over again

Author: Ahlström, Christer

Source: Linköping University

Download URL 2: Visit Now

Leave a Comment