Classification of LFPs Signals in Autistic and Normal Mice Based on Convolutional Neural Network

In view of the time-consuming and expensive problems that traditional methods is used to diagnose Autism Spectrum Disorder (ASD) through scales or magnetic resonance images, this paper proposes a classification method for ASD based on intrusive Electroencephalogram signals, which assists in determining whether there is ASD. Using artificially induced ASD mice’ local field potentials (LFPs) data set and normal mice’ LFPs data set, this thesis takes advantage of statistical analysis methods to test the significance between the two data sets, designs an architecture based on one-dimensional convolutional neural network, and trains two classification model of LFPs through Adam algorithm. The experimental result shows there is a strong significant difference between the two comparison groups (P < 0.001). After multiple tests on the test set, the average classification accuracy is 99.05%, indicating that the analysis method based on LFPs is an effective auxiliary method for judging whether there is ASD.