ECG Shock Detection During CPR Using CNN Spectrogram Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing electrocardiographic analysis systems for determining the need for an electric shock during CPR have diagnostic accuracies of 95-98%, which are not sufficient for widespread implementation in AEDs, and there is a need for a system that can accurately determine the need for an electric shock virtually continuously without interrupting chest compressions.
Innovation Solution
An electrocardiographic analysis system that utilizes a convolutional neural network (CNN) to analyze ECG and transthoracic impedance spectrograms, generated from digitally sampled ECG and impedance signals, to determine the need for an electric shock, using self-learning optimization with sample data and short-time Fourier transform to achieve high accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If chest compressions are performed continuously for CPR, then the survival rate is improved, but the ECG signal becomes contaminated with noise making accurate diagnosis difficult
Solution Approach 1:
The patent introduces an intermediary noise estimation mechanism that mediates between the chest compression process and ECG signal acquisition. By estimating noise characteristics during compressions and subtracting them from the raw ECG signal, the system enables continuous CPR while maintaining diagnostic accuracy. This intermediary processing layer allows both continuous compressions and clean ECG analysis to coexist.
Solution Approach 2:
The system performs preliminary noise estimation and subtraction before ECG analysis during chest compressions. By pre-processing the ECG signal to remove compression-related noise artifacts in advance, the system prepares clean signal data for subsequent diagnostic algorithms, enabling accurate rhythm detection without interrupting CPR.
2Measurement precision
If ECG diagnosis is performed by visual inspection or automatic analysis, then the need for electric shock can be determined, but chest compressions must be interrupted for several seconds to 10 seconds or more
Solution Approach 1:
The patent implements continuous ECG monitoring and analysis throughout CPR without interruptions. By maintaining continuous signal acquisition and processing, the system eliminates the need to pause compressions for rhythm checks. The useful actions of CPR and ECG diagnosis occur simultaneously and continuously, maximizing both circulation support and diagnostic monitoring.
Solution Approach 2:
The system replaces the mechanical interruption of chest compressions with a digital signal processing approach. Instead of physically stopping compressions to obtain a clean ECG signal, the patent uses computational methods to filter and analyze the signal during ongoing compressions, substituting mechanical disruption with electronic processing.
3Measurement precision
If noise filtering is applied to remove compression artifacts, then the ECG signal quality is improved, but the diagnostic accuracy remains limited to 95-98%
Solution Approach 1:
The patent combines multiple signal processing techniques and data sources to create a composite diagnostic approach. By integrating noise estimation, spectral analysis, and multiple ECG lead information, the system creates a more robust diagnostic framework that exceeds the accuracy of individual methods, achieving greater than 98% diagnostic accuracy through composite analysis.
Solution Approach 2:
The system transitions from analyzing only the time-domain ECG signal to incorporating frequency-domain spectral analysis. By examining the ECG signal in both time and frequency dimensions, the system gains additional diagnostic information that improves accuracy. This multi-dimensional analysis approach allows for more reliable differentiation between shockable and non-shockable rhythms.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The system achieves a diagnostic accuracy of 99.5% or higher for determining the need for an electric shock during CPR, allowing virtually continuous and real-time determination without interrupting chest compressions, using ECG and impedance spectrograms processed by a CNN.
Implementation Method 1
a convolutional neural network (CNN) that outputs a result of determining the need for an electric shock with respect to the ECG spectrogram and the impedance spectrogram input to the convolutional neural network
Implementation Method 2
the ECG spectrogram transforming means is configured to perform a short-time Fourier transform by dividing ECG discrete data within a determination window of a prespecified duration into small segments each having a specific duration with a fixed time difference
Data Source
Figure 1
Figure 2~3
Figure 4-1
AI summary
To provide an electrocardiographic analysis system capable of determining the need for an electric shock to a patient undergoing cardiopulmonary resuscitation (CPR) with a higher accuracy. An electrocardiographic analysis system includes electrocardiogram (ECG) signal acquiring means 11, ECG signal sampling means 12, ECG spectrogram transforming means 13, impedance signal acquiring means 21, impedance signal sampling means 22, impedance spectrogram transforming means 23, a convolutional neural network (CNN) 4 including an input layer 41, an output layer 4O, sample data accumulation means 4L, and sample data input means 4T, and electric shock indication reporting means 5. The CNN is a priori provided with sample data including sample ECG spectrograms and sample impedance spectrograms obtained from a large number of subjects, and sample response data on the need for an electric shock, and is optimized by self-learning the sample data. [Selected Drawing] Fig. 1