Heart Sound Retrieval Using Non-Rigid Alignment and Dynamic Time Warping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for analyzing heart sounds and ECGs in cardiovascular disease diagnosis are limited in accurately matching heart sounds for disease identification, as they fail to effectively account for fine morphological variability and physiological differences, leading to errors in disease-specific interpretation.
Innovation Solution
A computer-implemented method for detecting audio similarity of heart sounds through pre-processing, including selecting a time duration, constructing a line segment approximation, defining an audio envelope, isolating fiducial points, and using non-rigid alignment transforms based on shape-based dynamic time warping to match heart sounds with a database, determining a measure of shape similarity, and ranking matches.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional feature extraction and classification methods are used for heart sound analysis, then the analysis process is relatively simple, but the accuracy of matching heart sounds and detecting fine morphological variability is insufficient
Solution Approach 1:
The heart sound signal is segmented into multiple phases (S1, S2, S3, S4) with fiducial points identified in each phase. This segmentation allows for precise localization and comparison of morphological features across different heart sounds, improving matching accuracy while maintaining manageable complexity through structured organization
Solution Approach 2:
The patent transforms the heart sound analysis from traditional time-domain feature extraction to a multi-dimensional approach by constructing audio envelopes, identifying fiducial points in both time and amplitude dimensions, and applying dynamic time warping. This dimensional transformation enables capturing fine morphological variability that conventional single-dimensional methods miss
2Adaptability or versatility
If parametric modeling methods are used to spot major disease differences, then major disease categories can be identified, but fine morphological variability due to heart rate and physiological differences cannot be accounted for
Solution Approach 1:
The patent employs dynamic time warping (DTW) to align heart sounds with varying heart rates and temporal structures. The DTW algorithm dynamically adjusts the time axis to match corresponding fiducial points across different heart sounds, making the system adaptable to physiological variations while preserving fine morphological details for precise comparison
Solution Approach 2:
The patent changes the analysis parameters from fixed parametric features to variable fiducial point positions and audio envelope shapes. By tracking the positions, amplitudes, and temporal relationships of fiducial points across different heart sounds, the system adapts to physiological differences while maintaining precision in detecting morphological variability
3Ease of operation
If rule-based methods with P-Q-R-S-T segment detection are used for ECG analysis, then disease-specific interpretation can be performed, but errors in estimation of feature values cause major errors in diagnosis
Solution Approach 1:
The patent introduces audio envelopes as an intermediary representation between the raw heart sound signal and the classification process. The envelopes smooth out noise and highlight the overall shape characteristics, serving as a robust mediator that preserves diagnostic information while reducing sensitivity to minor measurement errors in fiducial point detection
Solution Approach 2:
The patent creates simplified copies of heart sounds in the form of audio envelopes that capture the essential morphological features. These envelope copies serve as robust representations for comparison and classification, maintaining the diagnostic information needed for disease interpretation while being less susceptible to errors than direct analysis of raw signals or precise fiducial point measurements
Data Source
AI summary
The present invention exploits a visual rendering of heart sounds and models the morphological variations of audio envelopes through a constrained non-rigid translation transform. Similar heart sounds are then retrieved by recovering the corresponding alignment transform using a variant of shape-based dynamic time warping.