Genetic algorithm designed for optimization of neural network architectures for intracranial EEG recordings analysis.

The visual inspection of intracranial electroencephalography (iEEG) recordings is a fundamental diagnostic step conducted prior to epileptic surgery in patients with drug-resistant epilepsy. The iEEG measurement usually spans several days to weeks, and the manual inspection is very time-consuming. In recent years, deep neural networks were shown to be a promising tool for iEEG analysis, which might significantly improve the iEEG inspection process by preselecting segments with high clinical relevance e.g. pathological activity while ignoring artifacts. Here we present a genetic algorithm, which optimizes neural network architecture and signal pre-processing parameters for iEEG classification. Our method improved the macro F1 score of the state-of-the-art model in two independent datasets, from St. Anne's University Hospital (Brno, Czech Republic) and Mayo Clinic (Rochester, MN, USA), from 0.9076 to 0.9673 and from 0.9222 to 0.9400 respectively. The proposed method achieved significantly improved results when compared to the state-of-the-art benchmark model (McNemar's test, p<<0.01). The results indicate that machine-designed neural network architectures outperform architectures designed by the heuristic approach of a human expert. Furthermore, we show that well-designed data preprocessing significantly affects model performance.&#xD.