EEG Neural Network Training Using Automatic Motion Labeling

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing neural networks face challenges in accurately processing EEG data for brain computer interfaces due to the difficulty in obtaining suitable training data and the dependence on the neural network and training data quality.

Innovation Solution

A method is provided that involves obtaining concurrent EEG and movement reference data while a subject performs or imagines motions, using automatic classification and labeling to train a neural network, particularly functional neural networks (FNNs) or compact convolutional networks like EEGNet, for predicting body part motions, with optional motion sensor data or expected motion data from software state, to create a brain computer interface.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If neural networks are used to process EEG data for brain computer interfaces, then prediction capability is improved, but training data quality and availability become limiting factors

Engineering Contradiction:
Improveprediction capabilityVSAvoidtraining data quality
Core Design Contradiction:
ReliabilityVSLoss of information

Solution Approach 1:

The patent introduces movement reference data as an intermediary element that bridges EEG data and meaningful predictions. Movement reference data serves as a mediator that captures actual motion information, which then labels EEG data to create high-quality training pairs, resolving the limitation of having raw EEG data without direct semantic meaning for training purposes

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent applies preliminary classification to movement reference data before using it to label EEG data. By pre-processing and classifying movement reference data into meaningful categories (e.g., stationary, moving, direction), the system prepares high-quality labels in advance that can be directly used to train neural networks, improving training efficiency and data quality

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If manual labeling of EEG data is performed, then training data accuracy is improved, but time consumption and complexity increase

Engineering Contradiction:
Improvetraining data accuracyVSAvoidlabeling time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent implements an automatic labeling system where movement reference data self-generates labels for EEG data through classification. The system uses the subject's own movement information (captured via sensors or software state) to automatically create ground truth labels, eliminating the need for external manual annotation while maintaining high accuracy

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent replaces the mechanical process of manual labeling with an automated computational system. Instead of human experts manually annotating EEG data, the system uses automatic classification algorithms that process movement reference data and generate labels computationally, dramatically reducing time and human resource requirements

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentEP4650916A1Method for providing a neural network for analyzing EEG data for use in providing a brain computer interface
Publication Date: 2025.11.19 THE SENSOR BASED ADAPTIVE NEURAL PROSTHETICS GMBH
  • EP4650916A1 patent drawingFigure 1~2
  • EP4650916A1 patent drawingFigure 3
  • EP4650916A1 patent drawingFigure 4

AI summary

The invention provides a computer-implemented method for providing a neural network for analyzing EEG data for use in providing a brain computer interface, the method comprising obtaining training data from EEG data of a subject and movement reference data associated with the EEG data, the EEG data and movement reference data acquired concurrently and while a subject carries out motions and/or imagines motions of a body part directed at controlling a software, wherein obtaining the training data comprises automatic classification of the movement reference data and automatically labeling at least part of the EEG data based on the classification of the movement reference data associated with the EEG data; and using the training data to train a neural network to predict, from EEG data, a motion of the body part imagined and/or carried out by the subject.