Video Frame Stacking for Motor Seizure Detection and Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods fail to effectively monitor and classify seizures, particularly nocturnal seizures, which are often unnoticed, leading to the need for improved patient surveillance systems.
Innovation Solution
A system utilizing video, audio, and depth data analysis with pre-trained neural networks to detect anomalies and classify seizure types, including video frame stacking, feature extraction, and classification using deep neural networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional seizure monitoring methods (patient diaries, manual surveillance) are used, then implementation simplicity is maintained, but detection reliability and completeness deteriorate due to missed seizures especially during nighttime
Solution Approach 1:
The patent replaces manual surveillance and patient self-reporting with an automated computer vision system using deep neural networks to analyze video data. The system automatically detects seizures by processing video frames through trained neural networks, eliminating the need for continuous human monitoring and improving detection reliability without requiring complex medical equipment
Solution Approach 2:
The system creates a digital model (neural network) trained on labeled seizure data that can automatically recognize and classify seizure patterns. This copied knowledge from training data enables the system to reliably detect seizures without direct human intervention, resolving the contradiction between reliability and complexity
2Measurement precision
If multiple sensors and electrodes are used for seizure detection, then measurement precision improves, but ease of operation and patient comfort deteriorate due to inconvenient sensors
Solution Approach 1:
The patent substitutes physical sensors and electrodes with a non-contact video-based monitoring system. The deep neural network analyzes visual patterns in video data to detect and classify seizures, achieving high measurement precision without requiring any physical contact with the patient, thereby dramatically improving ease of operation and patient comfort
Solution Approach 2:
The system uses video data as an intermediary medium to capture seizure information without direct physical interaction. The neural network processes this intermediate visual information to infer seizure characteristics, enabling precise measurement while maintaining patient convenience
3Measurement precision
If video data analysis with neural networks is implemented, then seizure detection accuracy improves, but use of energy and computational resources increases
Solution Approach 1:
The system performs preliminary action by pre-training neural networks offline on large datasets before deployment. This advance preparation allows the model to be optimized and stored for efficient inference, reducing real-time computational energy requirements while maintaining high detection accuracy during actual seizure monitoring
Solution Approach 2:
The patent segments the computational task into distinct stages: video frame extraction, neural network inference, and seizure classification. This segmentation allows for optimized processing at each stage, reducing overall energy consumption while maintaining high measurement precision through specialized processing for each function
Data Source
AI summary
The invention relates to a method for detecting and classifying a motor seizure. The method comprises receiving video data of a patient; detecting a first anomaly in the video data as anomaly in movement of the patient; determining a video frame stack comprising the first anomaly; classifying the video frame stack using a pre-trained neural network to obtain a first classification; and determining a motor seizure type based on the first classification.