SeizNow: A Highly Accurate Model for Automated Pediatric Seizure Diagnosis Using Neural Networks

Abstract

The purpose of this experiment is to examine the mathematical relationships that exist within electroencephalogram (EEG) data and to determine if those relationships can help diagnose pediatric seizures. There is currently no cost-efficient automated way to detect seizures solely based on EEG data. This research aims to find a reliable way to detect seizures by determining the relationship between the discrete energy differences of pediatric seizure EEG data and pediatric non-seizure EEG data. A neural network was coded to analyze the discrete energy differences and classify EEG data into seizure and non-seizure groups with an 87% accuracy. The discrete energy differences between the two groups were statistically different, so the null hypothesis was rejected. Non-seizure data had higher discrete energy values, indicating that the EEG graphs for the seizure group's discrete energy differences were smoother overall and had less distinct changes between the relative minimums and maximums. This project highlights the importance of examining mathematical properties, such as discrete energy, to indicate seizure activity. This project was the first to successfully apply discrete energy to medical research. This neural network could be expanded to help automate seizure diagnosis in pediatric patients in real-time, assisting healthcare professionals. Notably, since the project was developed at no cost, it can also be applied to low-income communities, ensuring broader access to essential healthcare technology.