A convolutional neural network based framework for classification of seizure types.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Epileptic seizures stem from brain electrical disturbances and are classified using EEG and other parameters.
• Existing research primarily focuses on classifying seizures versus non-seizures, leaving seizure type classification underexplored.

Purpose of the Study:

• To propose and evaluate a convolutional neural network (CNN) based framework for the 8-class classification of epileptic seizure types.
• To address the gap in research regarding the computational classification of specific epileptic seizure types.

Main Methods:

• EEG time series data were transformed into spectrogram stacks.
• These spectrograms were utilized as input for several CNN models, including AlexNet, VGG16, VGG19, and a basic CNN.
• An 8-class classification problem was formulated to distinguish between seizure types and non-seizures.

Main Results:

• The proposed CNN framework achieved classification accuracies of 84.06% (AlexNet), 79.71% (VGG16), 76.81% (VGG19), and 82.14% (basic CNN).
• This study represents the first known application of a computational algorithm for classifying diverse seizure types.

Conclusions:

• The developed CNN-based framework demonstrates a viable approach for the accurate classification of epileptic seizure types.
• The findings suggest that this computational method can be a valuable tool for the neurology community in recognizing different seizure types.