Journal: Scientific Reports

Article Title: Design of MRI structured spiking neural networks and learning algorithms for personalized modelling, analysis, and prediction of EEG signals

doi: 10.1038/s41598-021-90029-5

Figure Lengend Snippet: The blue line is the wave form of the real input signal after filtering for comparative analysis; the red line is the predicted output of the MRI-cSNNr and the green line—the predicted output of the MRI-sSNNr models: (a) l earning and testing of EEG data in the proposed MRI-SNNr models—MRI-cSNNr and MRI-sSNNr are compared with the 15 input EEG channel data after 0-70 Hz MOEF filtering ; (b) learning and testing of EEG signals in the two proposed MRI-SNNr models for 15 input channels using 0–70 Hz output Kaiser filter compared with the filtered input EEG signals. The learning process of the Cz channel in (b) failed; although Fp1 is not accurately learned in (a) , the experiment in (a) outperforms (b) . (c) Predicted signals at 10 neurons that correspond to unmonitored signals not used for training, while the models were trained on the measured EEG data from the 15 channels. It is seen that the prediction accuracy within the MRI-cSNNr structure is higher.

Article Snippet: For a plausible comparison of the outputs of the models, both the input EEG signals and the network outputs are filtered simultaneously using a MATLAB–designed-Minimum-Ordered-Equripple-FIR filter (MOEF) with a 0–70 Hz pass frequency and 80 dB attenuation high frequency, containing the main frequency bands and eliminating drift (all filter parameters are listed in the ) .

Techniques:

Journal: Scientific Reports

Article Title: Design of MRI structured spiking neural networks and learning algorithms for personalized modelling, analysis, and prediction of EEG signals

doi: 10.1038/s41598-021-90029-5

Figure Lengend Snippet: MSE,SD and P-Value Test Comparison of Several Methods.

Article Snippet: For a plausible comparison of the outputs of the models, both the input EEG signals and the network outputs are filtered simultaneously using a MATLAB–designed-Minimum-Ordered-Equripple-FIR filter (MOEF) with a 0–70 Hz pass frequency and 80 dB attenuation high frequency, containing the main frequency bands and eliminating drift (all filter parameters are listed in the ) .

Techniques: Comparison