Seizure Forecasting via Weighted Probabilistic Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current seizure forecasting methods suffer from poor generalizability due to short historical data duration and rely on categorical assessments, which are inappropriate for probabilistic forecasts, limiting their accuracy and practicality in predicting seizure likelihood in epilepsy patients.
Innovation Solution
A method and system that generate a temporal probability model based on historical seizure data, combining it with physiological data from EEG signals and environmental variables to create a weighted probabilistic model for estimating seizure probability, providing a continuous probability forecast rather than categorical predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional categorical assessment methods are used for seizure prediction, then the evaluation framework is simple, but the accuracy and appropriateness for probabilistic forecasts deteriorates
Solution Approach 1:
The patent transitions from categorical parameters (seizure will or will not happen) to probabilistic parameters (seizure likelihood between 0 and 1). This parameter change enables more nuanced prediction accuracy by capturing uncertainty and varying degrees of risk, directly improving measurement precision for seizure prediction while requiring more complex probabilistic modeling frameworks.
2Reliability
If short historical data duration is used for forecasting, then the data processing time is reduced, but the generalizability and reliability of seizure prediction deteriorates
Solution Approach 1:
The patent implements a two-stage approach where long-term historical data is collected and processed in advance to train probabilistic models, then these pre-trained models are deployed for real-time prediction. This preliminary action with long-term data improves generalizability while minimizing the time loss during actual forecasting operations.
Solution Approach 2:
The system dynamically adapts the length of historical data windows based on patient-specific seizure patterns and model performance. For patients with stable patterns, shorter windows suffice, while those with variable patterns utilize longer historical contexts, optimizing the balance between reliability and time efficiency for each individual case.
3Measurement precision
If patient-specific long-term data is used for modeling, then the prediction accuracy is improved, but the model complexity and computational requirements increases
Solution Approach 1:
The patent segments the probabilistic modeling process into distinct modules: temporal pattern recognition, spectral analysis, and integrative prediction components. Each module processes specific aspects of patient data independently, then combines results to achieve high prediction accuracy. This segmentation reduces overall model complexity by breaking down the complex task into manageable, specialized sub-tasks that can be optimized separately.
Data Source
AI summary
A method of estimating the probability of a seizure in a subject, the method comprising: receiving historical data associated with epileptic events experienced by the subject over a first time period, the historical data comprising physiological data associated with each epileptic event and a time at which each epileptic event occurred; generating a temporal probability model of future epileptic events based on the time of each of the epileptic events, the temporal probability model representing a probability of a future seizure occurrence in each of a plurality of time windows; generating a probabilistic model based on the physiological data associated with each epileptic event; weighting the probabilistic model based on the temporal probability model to generate a weighted probabilistic model of future seizure activity; and outputting an estimate of seizure probability in the subject using the weighted probabilistic model.