Ensemble Machine Learning for Bio-Impedance Respiratory Event Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting respiratory events, such as sleep apnea, using bio-impedance measurements are limited by the need for specific acquisition techniques and may not reliably detect events without processing human-engineered features, which can miss important markers.
Innovation Solution
A method employing an ensemble of machine learning models, including those with Long-Short Term Memory (LSTM) cells, to analyze bio-impedance measurement signals without pre-processing for human-engineered features, allowing for robust detection of respiratory events by comparing signals against unique training data sets and potentially combining with additional measurement signals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple separate bio-impedance measurements are used (trans-thoracic and trans-cervical), then detection reliability improves, but device complexity and measurement requirements worsen
Solution Approach 1:
The patent combines multiple separate bio-impedance measurement signals (trans-thoracic and trans-cervical) into a single integrated analysis framework. The ensemble machine learning model processes multiple measurement channels simultaneously, merging the information from different anatomical locations to detect respiratory events reliably without requiring separate dedicated measurement systems.
Solution Approach 2:
The invention creates a universal detection system that can process various types of bio-impedance measurement signals from different locations (thorax, neck, etc.) using the same machine learning ensemble framework. This multi-functional approach allows the system to handle diverse measurement configurations without requiring location-specific processing algorithms.
2Loss of information
If human-engineered feature processing is used, then measurement interpretation improves, but important respiratory markers may be missed
Solution Approach 1:
The patent extracts and removes the step of manual human-engineered feature processing from the detection pipeline. Instead of pre-defining features through human expertise, the system directly feeds raw bio-impedance measurement signals into the machine learning ensemble, which automatically learns and extracts relevant features, thereby preserving all original signal information including subtle respiratory markers.
Solution Approach 2:
The invention replaces the mechanical approach of manual feature engineering with an intelligent machine learning system. The ensemble of models automatically performs feature extraction and pattern recognition, substituting human cognitive processing with computational algorithms that can identify subtle respiratory markers without manual intervention.
3Reliability
If ensemble machine learning models are used, then detection robustness improves, but computational requirements worsen
Solution Approach 1:
The patent performs preliminary actions by pre-training the ensemble machine learning models offline using extensive training datasets. This preliminary training phase captures the computational intensity, while the deployed system only requires inference operations. The models are prepared in advance to handle the complexity of ensemble processing, reducing real-time computational energy requirements.
Solution Approach 2:
The system dynamically adjusts the ensemble model processing based on input signal characteristics. The machine learning ensemble can adaptively weight different models or reduce computational complexity when signal conditions allow, optimizing energy consumption while maintaining detection robustness across varying respiratory conditions.
Data Source
AI summary
A method for detecting a respiratory event of a subject comprises: receiving a bio-impedance measurement signal (S2) dependent on respiratory action from the subject; extracting (306) at least one time-sequence of the bio-impedance measurement signal (S2); and for each extracted time-sequence: comparing (308) the bio-impedance measurement signal (S2) with each of a plurality of machine learning models in an ensemble of machine learning models to form a set of predictions of occurrence of a respiratory event, wherein each prediction is based on comparing the bio-impedance measurement signal (S2) with one machine learning model, wherein each model correlates features of time-sequences of a bio-impedance measurement signal (S2) with presence of a respiratory event and wherein each model is trained on a unique data set of training time-sequences; deciding (310) whether a respiratory event occurs in the extracted time-sequence based on the set of predictions.