Optimized Multilayer Perceptron for Sensorimotor Functional Mapping Based on a Few Minutes of Intracranial Electroencephalogram Data.

Abstract

Using 6-minute free-running intracranial-electroencephalogram (icEEG) during sleep, an optimized multilayer perceptron (MLP) neural network accurately maps the sensorimotor cortex (SM) and identifies the anterior lip of the central sulcus (CS) in intractable epilepsy patients. We calculated 6 performance metrics to evaluate the MLP's efficacy: accuracy, area under the curve (AUC), recall, precision, F1-scores, and specificity. Each layer had 4 neurons with hyperbolic TanH activation function and 4 with Gaussian distribution function. Conventional 10-fold cross-validation was used. Feature extension (ε) and weighted imbalanced data (w) improved MLP performance. ANN NEUROL 2024;96:187-193.