AMI Alert Accuracy in Wearable ECG via Reservoir Computing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies for detecting acute myocardial infarction (AMI) using non-medical wearable devices face challenges in reducing the time required to initiate alerts and improving the accuracy and reliability of these alerts, particularly due to limitations in acquiring and analyzing multi-lead ECGs.
Innovation Solution
A method that compensates for the reduced number of ECG leads by processing physiological signals, extracting static and dynamic AMI features, and using a reservoir computing model, such as an echo state neural network, to generate alerts for suspected AMI based on a predefined threshold.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a standard 12-lead ECG is used for AMI detection, then measurement precision is improved, but device complexity and time required for acquisition increase
Solution Approach 1:
The patent segments the 12-lead ECG into multiple single-lead ECGs that can be acquired sequentially using a simplified wearable device configuration. This allows the system to reconstruct comprehensive cardiac information through time-multiplexed acquisition rather than requiring all 12 leads simultaneously, thereby reducing device complexity while maintaining measurement precision.
Solution Approach 2:
The patent transitions from spatial dimension (12 simultaneous leads) to temporal dimension by acquiring multiple single-lead ECGs at different time points and orientations. This dimensional transformation allows reconstruction of comprehensive cardiac electrical activity using a simpler device that acquires data sequentially rather than requiring complex simultaneous multi-lead configuration.
2Measurement precision
If a standard 12-lead ECG is used for AMI detection, then measurement precision is improved, but the time required to acquire and analyze the ECG increases
Solution Approach 1:
The patent implements preliminary action by continuously monitoring basic physiological parameters and pre-processing ECG data in real-time. When suspicious patterns are detected, the system is already prepared with pre-computed features and algorithms ready to rapidly analyze additional single-lead ECGs, significantly reducing the time from symptom onset to alert generation compared to waiting for a full 12-lead ECG.
Solution Approach 2:
The patent applies the skipping principle by implementing a streamlined acquisition protocol that obtains essential diagnostic information through a reduced set of single-lead ECGs rather than completing a full 12-lead acquisition. The system strategically selects and acquires only the most informative leads needed for rapid AMI detection, rushing through the essential diagnostic steps without unnecessary delays.
3Device complexity
If fewer ECG leads are used in wearable devices, then device complexity and acquisition time are reduced, but reliability of AMI alerts deteriorates
Solution Approach 1:
The patent introduces an intermediary processing system that acts as a mediator between the simplified single-lead ECG acquisition and reliable AMI detection. This intermediary layer includes advanced signal processing algorithms, feature extraction mechanisms, and machine learning models that compensate for the reduced lead count by intelligently analyzing and synthesizing diagnostic information from limited inputs, thereby maintaining alert reliability despite device simplicity.
Solution Approach 2:
The patent applies parameter changes by transforming the limited single-lead ECG signals through various signal processing operations, feature transformations, and temporal analysis methods. By changing the parameters of analysis (frequency domain transformation, time-frequency analysis, statistical features) rather than relying on increased spatial sampling through more leads, the system maintains detection reliability while using a simpler device configuration.
Data Source
Figure 1
Figure 2~2b
Figure 3
AI summary
This disclosure presents a method for reducing the time required to initiate an alert on a suspected acute myocardial infarction (AMI) and improving the accuracy and reliability of the alert. The method involves acquiring and processing physiological signals, including multi-lead ECG and PPG signals captured using a wrist-worn device equipped with biooptical sensors and biopotential electrodes. These signals are utilized to extract both static and dynamic AMI features, such as ST segment abnormality index, QRS-T angle, the occurrence of bundle branch block, ECG signal quality, reduction in oxygen saturation, the number of ventricular premature beats, the occurrence of ventricular arrhythmias, and PPG signal quality index. These extracted AMI features are then employed as inputs to train a reservoir computing model, specifically an echo state neural network, using individualized data. The model computes an alert score, and if it surpasses a predefined threshold, an alert for suspected AMI is generated.