Contrastive Learning Sleep Stage Classification Reducing User Performance Variation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automatic sleep stage classification systems face significant variation in performance across different users and environments, limiting their reliability and effectiveness in real-life applications, despite advancements in biosignal measurement technologies and deep learning.
Innovation Solution
An automatic sleep stage classification system utilizing a contrastive learning method that preprocesses and measures biosignals such as EEG, EOG, EMG, respiratory effort signals, pulse, and oxygen saturation, extracts unique features, and classifies sleep stages based on shared features among multiple users, reducing performance variation through a Siamese network structure and feature extraction units.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If EEG-based automatic sleep stage classification systems are used to achieve high performance in controlled environments, then classification accuracy is improved, but performance variation increases when operating in uncontrolled daily life situations
Solution Approach 1:
The patent applies parameter changes by transforming the input data from raw biosignals to standardized z-score normalized features. This normalization process adjusts the statistical parameters (mean and standard deviation) of the biosignal data, allowing the system to adapt to different users and environments while maintaining consistent classification performance. The feature extraction and classification parameters are optimized to work effectively across diverse conditions.
Solution Approach 2:
The patent implements feedback mechanisms through continuous performance monitoring and model retraining. The system collects classification results and compares them against ground truth labels, using this feedback to refine the machine learning models. This iterative feedback loop allows the system to learn from real-world performance variations and improve its accuracy over time, addressing the reliability issue in uncontrolled environments.
2Measurement precision
If deep learning models are trained on specific datasets to achieve high accuracy, then classification performance is improved, but the system fails to generalize to different users and environments
Solution Approach 1:
The patent applies universality by developing a multi-functional feature extraction system that can handle multiple biosignal types (EEG, EOG, EMG, respiratory effort, pulse, oxygen saturation) and adapt to different users. The system uses a universal feature extraction architecture that processes various biosignal modalities through common transformation layers, enabling it to maintain high accuracy across diverse users and environments without requiring user-specific retraining.
Solution Approach 2:
The patent implements preliminary action through pre-processing and feature extraction performed before classification. The system performs preliminary z-score normalization and feature transformation on the raw biosignal data, creating standardized feature representations that are ready for classification. This preliminary processing step prepares the data in a way that enhances generalization to different users and environments, reducing the need for user-specific model adaptation.
3Measurement precision
If user-specific models are trained to improve individual accuracy, then performance for specific users is improved, but the system becomes complex and difficult to deploy
Solution Approach 1:
The patent applies segmentation by dividing the sleep stage classification system into independent functional modules: biosignal acquisition, pre-processing, feature extraction, and classification. Each module operates independently and can be optimized separately. The feature extraction module segments the biosignal processing into distinct transformation stages (z-score normalization, feature computation), making the system more manageable and easier to deploy while maintaining user-specific accuracy through modular customization.
Data Source
AI summary
An automatic sleep stage classification system for reducing the variation in performance between users using a contrastive learning method according to one embodiment of the present invention includes: a user terminal that measures a user's biosignal and preprocesses the user's measured biosignal; and a classification server that receives the user's preprocessed biosignal from the user terminal, extracts the user's unique biosignal feature, extracts a similar feature by comparing the user's extracted unique biosignal feature and the user's biosignal feature for contrastive learning, and classifies sleep stages based on the extracted similar feature, wherein the user's biosignal includes at least one of the user's electroencephalography (EEG), electrooculography (EOG), electrocardiogramalectromyography (EMG), respiratory effort signals, pulse, oxygen saturation (SpO2), and blood flow.