ECG Noise Discrimination via PSD Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current cardiac monitoring devices, including wearable defibrillators and implantable devices, face challenges in accurately distinguishing between life-threatening arrhythmias and noise in electrocardiogram (ECG) signals, leading to false-positive detections and potential unnecessary treatments.
Innovation Solution
The implementation of a cardiac monitoring device that utilizes power spectral density (PSD) analysis and machine learning algorithms to differentiate between cardiac events and noise by extracting features such as dominant frequency, in-band entropy, first-band entropy, and variance from the ECG signal, and comparing these features to predetermined threshold scores to determine the likelihood of a cardiac event.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If high sensitivity detection is used to detect life-threatening arrhythmias, then detection sensitivity is improved, but false-positive detection rate increases
Solution Approach 1:
The detection process is divided into multiple independent stages: initial event detection, power spectral density analysis, feature extraction (dominant frequency, in-band entropy, first-band entropy, variance), and final classification. Each stage independently evaluates specific characteristics, allowing the system to maintain high sensitivity while filtering false positives through cumulative evidence assessment.
Solution Approach 2:
The patent transforms the ECG signal from time domain to frequency domain using power spectral density analysis. This dimensional transformation enables the system to evaluate signal characteristics in multiple frequency bands simultaneously, providing additional discrimination dimensions that distinguish true arrhythmias from noise while maintaining detection sensitivity.
2Measurement precision
If traditional noise filtering is applied to reduce false detections, then false-positive rate decreases, but detection sensitivity is reduced
Solution Approach 1:
Instead of applying uniform noise filtering across the entire signal, the patent extracts localized features from specific frequency bands (in-band entropy for 2-6Hz, first-band entropy for 0-2Hz). This localized feature extraction preserves important arrhythmia characteristics while filtering noise, achieving reduced false positives without sacrificing detection sensitivity.
Solution Approach 2:
The system changes the evaluation parameters from simple amplitude or rate-based metrics to power spectral density features including dominant frequency, in-band entropy, first-band entropy, and variance. These parameter transformations enable sophisticated noise discrimination while maintaining the ability to detect true arrhythmias, resolving the contradiction between false-positive reduction and sensitivity preservation.
3Measurement precision
If multiple feature extraction and machine learning algorithms are implemented, then detection accuracy is improved, but device complexity increases
Solution Approach 1:
The patent performs preliminary power spectral density transformation and feature extraction (dominant frequency, in-band entropy, first-band entropy, variance) before final classification. This preliminary processing organizes the signal data into meaningful features that simplify subsequent machine learning classification, achieving high detection accuracy while managing computational complexity through structured preprocessing.
Solution Approach 2:
The system creates a transformed representation (power spectral density) of the original ECG signal that captures essential characteristics in a different domain. This copied representation in the frequency domain serves as a simplified input for classification algorithms, maintaining detection accuracy while reducing the complexity of direct time-domain analysis.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A cardiac monitoring device includes: at least one sensing electrode for obtaining an electrocardiogram (ECG) signal from a patient; a processing unit comprising at least one processor operatively coupled to the at least one sensing electrode. The cardiac monitoring device is configured to: obtain the ECG signal from the at least one sensing electrode; determine a transformed ECG signal based on the ECG signal; extract at least one value representing at least one feature of the transformed ECG signal; provide the at least one value to determine a score associated with the ECG signal, thereby providing an ECG-derived score; compare the ECG-derived score to a predetermined threshold score determined by machine learning; and provide an indication of a cardiac event if the ECG-derived score is one of above or below the predetermined threshold score determined by the machine learning.