SeizyML: An Application for Semi-Automated Seizure Detection Using Interpretable Machine Learning Models.

Area of Science:

• Neuroscience
• Computational Biology
• Bioinformatics

Background:

• Accurate seizure detection from electrographic recordings is crucial for epilepsy research.
• Current manual analysis methods are time-consuming, inefficient, and prone to bias.
• A lack of open-source automated tools hinders progress in the field.

Purpose of the Study:

• To develop and validate SeizyML, an open-source software for automated electrographic seizure detection.
• To compare the performance of interpretable machine learning classifiers for seizure detection.
• To provide an efficient and accurate tool for epilepsy research.

Main Methods:

• Development of SeizyML, an open-source software integrating machine learning with manual validation.
• Comparison of four interpretable machine learning classifiers: decision tree, Gaussian Naive Bayes, passive aggressive classifier, and stochastic gradient descent.
• Training and validation on an extensive electrographic seizure dataset from chronically epileptic mice.

Main Results:

• The Gaussian Naive Bayes model demonstrated superior performance, detecting all seizures.
• This model exhibited the lowest false detection rate and robustness to misclassifications.
• Effective seizure detection was achieved with a minimal amount of training data.

Conclusions:

• SeizyML offers a transformative solution for the analysis bottleneck in epilepsy research.
• The Gaussian Naive Bayes classifier within SeizyML provides efficient, accurate, and unbiased seizure detection.
• This open-source tool has the potential to accelerate research progress in understanding and treating epilepsy.