Seizure Forecasting Using Transfer Learning on Subcutaneous EEG

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current methods for predicting seizure onset in epilepsy patients using subcutaneous EEG data face challenges such as limited data availability and high processing requirements for deep learning algorithms, which hinder accurate and efficient forecasting.

Innovation Solution

A multi-stage training process using a trained machine learning algorithm, specifically a three-layer long-short-term-memory (LSTM) neural network, is employed to classify subcutaneous EEG data, leveraging transfer learning from scalp-recorded EEG data to improve seizure prediction accuracy and reduce training time.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If deep learning algorithms are used for seizure prediction, then prediction accuracy is improved, but processing requirements and training time increase significantly

Engineering Contradiction:

Improveseizure prediction accuracyVSAvoidprocessing requirements

Core Design Contradiction:

Measurement precisionVSUse of energy by stationary object

Solution Approach 1:

The patent uses transfer learning where a deep learning model pre-trained on scalp EEG data is copied and adapted for subcutaneous EEG data. The pre-trained model's weights are initialized and then fine-tuned on the smaller subcutaneous EEG dataset, allowing the system to leverage the computational resources already invested in training the deep learning architecture on more abundant scalp EEG data, thereby reducing the processing requirements for training on subcutaneous data alone

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The deep learning algorithm performs preliminary training on scalp-recorded EEG data before being applied to subcutaneous EEG data. This preliminary action establishes the foundational model capabilities in advance, so that when the model is later applied to subcutaneous EEG data for seizure prediction, it requires significantly less computational resources and training time compared to training from scratch

Inventive Principle:

Principle #10Preliminary action

2Measurement precision

If deep learning algorithms are used for seizure prediction, then prediction accuracy is improved, but training time increases significantly

Engineering Contradiction:

Improveseizure prediction accuracyVSAvoidtraining time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent applies transfer learning by copying the pre-trained deep learning model parameters from scalp EEG training and adapting them for subcutaneous EEG data. This copying approach allows the model to start with already-learned features and patterns from the larger scalp EEG dataset, significantly reducing the training time required for subcutaneous EEG data while maintaining high prediction accuracy

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The system performs preliminary training on scalp EEG data to establish the deep learning model before applying it to subcutaneous EEG data. This preliminary action completes the heavy computational work in advance, so that subsequent training on subcutaneous data only requires fine-tuning, dramatically reducing the overall training time for the specific application

Inventive Principle:

Principle #10Preliminary action

3Duration of action of stationary object

If subcutaneous EEG data is used for seizure prediction, then long-term monitoring capability is improved, but data availability for training is limited

Engineering Contradiction:

Improvelong-term monitoring capabilityVSAvoidtraining data availability

Core Design Contradiction:

Duration of action of stationary objectVSQuantity of substance

Solution Approach 1:

The patent makes the deep learning model universal by first training it on the abundant scalp EEG data from multiple sources, creating a general-purpose seizure detection model. This universal model can then be adapted to work with subcutaneous EEG data from individual patients, allowing the same model architecture and training methodology to serve both the general population and specific patient monitoring applications

Inventive Principle:

Principle #6Universality (Multi-functionality)

Solution Approach 2:

The system copies the pre-trained deep learning model parameters from the general scalp EEG training data and adapts them to the specific subcutaneous EEG data from individual patients. This copying approach allows the model to leverage the extensive training data from scalp EEG while applying it to the limited subcutaneous EEG data available for each patient, effectively overcoming the data scarcity problem

Inventive Principle:

Principle #26Copying

Data Source

PatentUS20240315648A1Seizure Forecasting in Subsutaneous Electroencephalography Data Using Machine Learning

Publication Date: 2024.09.26 MAYO FOUNDATION FOR MEDICAL EDUCATION & RESEARCH

AI summary

Seizure onset can be forecast in subjects from subcutaneous electroencephalography (EEG) data input to a trained machine learning algorithm. A multi-stage training process is used to train the machine learning algorithm. A first stage of the training process is implemented on scalp-recorded EEG data. A second stage of the training process may be implemented on subcutaneously recorded EEG data.