Semi-automated seizure detection using interpretable machine learning models.

Area of Science:

• Neuroscience
• Computational Biology
• Biomedical Engineering

Background:

• Automated electrographic seizure detection is lacking validated open-source tools.
• Manual seizure curation is labor-intensive, inefficient, error-prone, and biased.
• This limits research progress in epilepsy and related neurological disorders.

Purpose of the Study:

• To develop and validate an open-source software (SeizyML) for automated electrographic seizure detection.
• To compare the performance of interpretable machine learning models for seizure detection.
• To demonstrate the utility of SeizyML in both animal models and human EEG data.

Main Methods:

• Developed SeizyML, an open-source software integrating machine learning with manual validation.
• Compared four interpretable machine learning models: decision tree, Gaussian Naive Bayes, Passive Aggressive Classifier, and Stochastic Gradient Descent.
• Trained and validated models on an electrographic seizure dataset from chronically epileptic mice and a human EEG dataset.

Main Results:

• Gaussian Naive Bayes and Stochastic Gradient Descent models demonstrated the highest precision and F1 scores.
• These models successfully detected all seizures in the mouse dataset.
• Effective seizure detection was achieved with minimal training data, indicating model efficiency.

Conclusions:

• SeizyML provides an efficient, accurate, and less biased method for electrographic seizure detection.
• The developed machine learning models, particularly Gaussian Naive Bayes and SGD, are effective for seizure analysis.
• This open-source tool has the potential to significantly accelerate research by overcoming current analysis bottlenecks.