Unsupervised ML Artifact Removal from EDA Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for artifact removal from electrodermal activity (EDA) data, particularly in clinical settings like surgery, are inadequate for real-time processing and often distort or fail to remove artifacts such as those caused by surgical cautery, limiting the use of EDA in clinical decision-making.
Innovation Solution
The implementation of unsupervised machine learning techniques, including isolation forest, K-nearest neighbor distance, and 1-class support vector machine, to identify and remove artifacts from EDA data by analyzing feature vectors and determining appropriate cut-offs, allowing for real-time artifact correction and preservation of true EDA data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional artifact removal methods (thresholding, filtering) are used, then processing speed is improved, but artifact removal accuracy deteriorates and true EDA data may be distorted or removed
Solution Approach 1:
The patent segments the EDA data into small windows (e.g., 1-second intervals) and processes each window independently using unsupervised learning algorithms. This segmentation allows real-time processing while maintaining high accuracy by analyzing local patterns rather than requiring global data processing.
Solution Approach 2:
The patent replaces conventional mechanical filtering methods (thresholding, high-pass filtering) with unsupervised machine learning algorithms (isolation forest, K-nearest neighbor, 1-class support vector machine). This substitution enables the system to automatically learn and distinguish artifact patterns from true EDA signals without manual parameter tuning, achieving both real-time processing and high accuracy.
2Measurement precision
If supervised learning tools are used for artifact removal, then artifact detection accuracy is improved, but implementation complexity and data preparation time increase
Solution Approach 1:
The patent employs unsupervised learning algorithms that automatically learn artifact patterns directly from the EDA data without requiring manual labeling or training datasets. The isolation forest, K-nearest neighbor, and 1-class support vector machine algorithms self-organize to distinguish artifacts from true signals based on statistical properties, eliminating the need for time-consuming data annotation and model training.
Solution Approach 2:
The patent changes the approach from supervised learning (requiring labeled data) to unsupervised learning (using unlabeled data with statistical parameter analysis). By transforming the problem into detecting statistical anomalies in segments of EDA data, the system achieves high detection accuracy without the complexity of supervised learning infrastructure.
3Object-generated harmful factors
If aggressive artifact removal is applied, then artifact elimination is improved, but loss of true EDA data increases
Solution Approach 1:
The patent applies local quality by analyzing small, localized segments of EDA data (e.g., 1-second windows) and making artifact removal decisions independently for each segment. This localized approach allows the system to preserve true EDA signals that occur between artifacts while removing artifact-contaminated segments, minimizing information loss while effectively eliminating artifacts.
Solution Approach 2:
The patent converts the presence of artifacts into a beneficial detection opportunity by using unsupervised learning to identify statistical anomalies. The artifacts, which create distinct statistical patterns in the EDA data, are transformed from harmful interference into detectable features that the algorithm can automatically identify and remove while preserving the underlying true signals.
4Measurement precision
If manual artifact detection is used, then detection accuracy is improved, but processing time and automation level deteriorate
Solution Approach 1:
The patent replaces manual visual inspection with automated unsupervised learning algorithms that process EDA data in real-time. The isolation forest, K-nearest neighbor, and 1-class support vector machine algorithms automatically detect artifacts by analyzing statistical properties of data segments, achieving manual-level accuracy without the time cost of human review.
Solution Approach 2:
The patent segments the continuous EDA data stream into manageable 1-second windows that can be processed independently and quickly by the unsupervised learning algorithms. This segmentation enables real-time automated processing while maintaining high detection accuracy through localized pattern recognition.
Data Source
AI summary
Systems and methods for identifying and removing artifacts from electrodermal activity (EDA) data are described herein. A method includes identifying artifacts in segments of EDA data using unsupervised machine learning based on feature vectors extracted from segments of the data. After the artifacts are identified, they can be removed from the EDA data. Artifact-free EDA data can be used to estimate a patient's nociceptive state, which in turn can be used to modify a dosage of anesthetic drugs administered to the patient based on this estimation.