EEG-Based Stroke Detection Using ML and Simulated Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current diagnostic techniques for stroke have difficulty accurately determining the type, location, and infarct volume/size of a stroke, leading to inadequate treatment and outcomes.
Innovation Solution
The use of an implantable or external medical device equipped with electrodes to sense electrical signals from the brain, generating EEG signals, and applying them to a machine learning model trained on both training EEG data and simulated EEG data to determine characteristics of a brain event.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If non-invasive techniques are used for stroke detection, then patient safety and comfort are improved, but the accuracy in determining stroke type, location, and infarct volume deteriorates
Solution Approach 1:
The patent uses EEG signals as an intermediary to indirectly detect stroke characteristics. Instead of directly imaging the brain (which would require invasive procedures), the system measures electrical activity on the scalp and uses machine learning to infer stroke type, location, and volume from these indirect measurements, thus maintaining non-invasiveness while improving detection accuracy
Solution Approach 2:
The system transforms the detection approach by changing from direct anatomical imaging parameters to functional electrical activity parameters. By analyzing EEG signal characteristics (frequency, amplitude, patterns) and applying machine learning models, the system can determine stroke features without the need for invasive imaging procedures
2Loss of time
If machine learning models are trained only on limited training EEG data, then training time and computational resources are reduced, but coverage gaps in brain event detection increase
Solution Approach 1:
The patent applies preliminary action by pre-training the machine learning model on simulated EEG data that covers a wide range of possible stroke scenarios before deployment. This pre-training ensures comprehensive coverage of different stroke types, locations, and volumes, so the model is ready to handle diverse real-world cases without requiring extensive retraining
Solution Approach 2:
The system creates synthetic copies of EEG data through simulation to augment the training dataset. By generating realistic simulated EEG signals representing various stroke conditions, the model learns from a diverse set of examples without requiring actual patient data for every possible scenario, thus improving coverage while maintaining efficient training
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach reduces coverage gaps in brain event detection, enabling more accurate identification of stroke types, locations, and infarct volumes, which can lead to more targeted and effective treatments.
Implementation Method 1
using electrodes, the medical device may sense electrical signals from a patient and generate EEG signal(s) based on the electrical signals
Data Source
AI summary
An example system includes a memory; a plurality of electrodes; sensing circuitry configured to: sense, via at least two electrodes of the plurality of electrodes, electrical signals from a patient; and generate, based on the electrical signals, one or more electroencephalography (EEG) signals; and processing circuitry configured to: receive, from the sensing circuitry, one or more EEG signals; and apply the one or more EEG signals to a machine learning (ML) model to determine one or more characteristics of a brain event, the ML model being trained on training EEG data and simulated EEG data.