Deep Convolutional Autoencoder for EDA Artifact Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for analyzing electrodermal activity (EDA) signals are hindered by motion artifacts, leading to unreliable data, especially in wearable technologies, and existing techniques for artifact reduction are inefficient and distort the skin conductance response.
Innovation Solution
A deep convolutional autoencoder network (DCAE) is applied to EDA signals to reduce artifacts, generating a modified signal that allows for accurate determination of health conditions, such as the risk of seizure or central nervous system oxygen toxicity, by performing time-invariant and time-variant spectral analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional time-domain methods are used to analyze EDA signals, then the analysis process is simple, but the measurement precision and reliability are low due to motion artifacts
Solution Approach 1:
The patent replaces conventional time-domain signal processing methods with frequency-domain spectral analysis. This substitution transforms the approach from simple temporal averaging to sophisticated spectral decomposition, thereby improving measurement precision by separating signal components based on their frequency characteristics rather than relying on time-domain features that are susceptible to motion artifacts.
Solution Approach 2:
The patent changes the analysis parameter from time-domain features (skin conductance level, phasic responses) to frequency-domain features (spectral power distribution, dominant frequencies). This parameter transformation enables the system to achieve higher measurement precision by analyzing the EDA signal in the frequency domain where motion artifacts can be distinguished from physiological signals through their different spectral signatures.
2Reliability
If artifact reduction techniques are applied to EDA signals, then the reliability of health condition determination improves, but the processing time and computational complexity increase
Solution Approach 1:
The patent applies artifact reduction techniques continuously in the background during EDA signal acquisition, rather than as a post-processing step. This preliminary action ensures that the spectral analysis always operates on cleaned signals, improving reliability without adding noticeable delay to health condition determination. The system prepares artifact-reduced signals in advance, making the actual health assessment faster and more reliable.
3Measurement precision
If deep convolutional autoencoder networks are used for artifact reduction, then the EDA signal quality improves significantly, but the computational power requirements increase
Solution Approach 1:
The patent performs the computationally intensive deep convolutional autoencoder processing in advance, during periods when the wearable device is being worn but not actively monitoring for seizures. By pre-processing EDA signals to remove artifacts before storage, the system reduces the computational burden during critical real-time monitoring phases, thereby lowering energy consumption when it matters most while still achieving high signal quality.
4Reliability
If motion artifacts are removed from EDA data, then the usability of the data improves, but the amount of usable data decreases when using conventional detection and discarding methods
Solution Approach 1:
The patent transforms the EDA signal from time-domain to frequency-domain representation through spectral analysis. This parameter change enables the system to identify and remove motion artifacts based on their frequency characteristics while preserving the physiological EDA signal components. Consequently, the system can maintain a higher quantity of usable data compared to conventional methods that discard entire segments containing artifacts, because frequency-domain analysis allows selective artifact removal without losing adjacent valid signal portions.
Data Source
AI summary
Methods, systems, non-transitory computer-readable media, and apparatuses are described for predicting a health condition of a subject. An apparatus may be configured to receive a physiological signal associated with the subject. The physiological signal may include artifacts. A modified physiological signal may be generated based on an application of a machine learning model to the physiological signal. The modified physiological signal may include the physiological signal with a reduction of the artifacts. A physiological measurement may be determined based on the modified physiological signal. The health condition may be determined based on a change in the physiological measurement satisfying a threshold. The apparatus may cause an output of an indication associated with the health condition.