PPG Signal Pulse Evaluation Using Morphological Comparison
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing PPG signal evaluation methods suffer from misclassifications due to noisy signals, adaptive template distortions, and reliance on expert labeling, leading to inaccurate physiological parameter estimation.
Innovation Solution
A method involving multiple predetermined pulse templates adapted to PPG signal conditions, with morphological comparisons and dynamic time warping to quantify pulse differences, rejecting bad quality pulses and detecting misclassifications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If adaptive templates are used to assess PPG signal quality, then the method can identify normal sinus rhythm beats, but the templates become distorted by noisy signals and arrhythmic conditions leading to false rejection of good pulses
Solution Approach 1:
The patent divides the pulse assessment into multiple independent quality indices (morphological quality index, amplitude quality index, temporal quality index) rather than using a single adaptive template. Each index evaluates specific aspects of pulse quality independently, preventing distortion propagation across all assessment dimensions.
Solution Approach 2:
The patent introduces reference pulse templates as intermediaries that represent ideal pulse morphology. These reference templates are compared against actual pulses through multiple quality indices, serving as a stable mediator that does not adapt to noisy inputs, thereby preventing template distortion while maintaining assessment accuracy.
2Measurement precision
If machine learning methods are used for PPG signal quality assessment, then the algorithm can distinguish between good and bad pulses, but the performance deteriorates when exposed to new datasets requiring periodic retraining
Solution Approach 1:
The patent implements self-service through the use of reference pulse templates that are pre-defined and do not require external training data or expert labeling. The system automatically assesses pulse quality by comparing against these fixed references, eliminating the need for periodic retraining while maintaining high accuracy across different datasets and populations.
Solution Approach 2:
The patent changes the fundamental parameter of pulse assessment from adaptive machine learning models to fixed reference templates. This parameter change transforms the system from one requiring continuous training to one that maintains stable performance across new datasets, as the reference templates remain constant regardless of population variations.
3Measurement precision
If expert labeling is used to train machine learning algorithms, then the algorithm can learn to distinguish good and bad pulses, but the labeling is not always accurate leading to misclassification
Solution Approach 1:
The patent uses copying by creating reference pulse templates that replicate ideal pulse morphology. Instead of relying on imperfect expert labels, the system copies the characteristics of normal pulses into reference templates, which then serve as the ground truth for assessment. This copying approach eliminates the inaccuracies inherent in expert labeling while maintaining the ability to differentiate pulse qualities.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
The present invention relates to a method (1) of evaluating a detected pulse in a Photoplethysmography, PPG, signal (10) in a PPG signal measurement, the method comprising the steps of (a) receiving the PPG signal comprising a plurality of pulses (20); (b) extracting pulses from the received PPG signal; (c) providing one or more predetermined pulse templates (30); (d) selecting one or more pulse templates (32) out of the provided pulse templates (30); (e) adapting the selected one or more pulse templates to the shape of a first pulse (22) out of the identified pulses (20) in the PPG signal (10); and (f) performing a morphological comparison between the first pulse (22) and each of the adapted one or more pulse templates (34) to provide a quantification of the difference (AD) between the first pulse and the adapted one or more pulse templates.