Multimodal Feature Extraction Algorithm for Potential Frequency in the Fusion of EEG and MRI

To address the problem that the traditional Electroencephalogram (EEG) induces insufficient consideration of multimodal in the potential frequency feature algorithm leading to the degradation of data quality, which in turn affects the accuracy of potential frequency feature extraction. Therefore, we propose a potential frequency multimodal feature extraction algorithm in the Fusion of EEG and Magnetic Resonance Imaging (MRI). First, the multimodal EEG was preprocessed. The Empirical mode decomposition was used to decompose the multimodal EEG into multiple Intrinsic Mode Function (IMF) components, extract the principal IMF components, and complete the reconstruction of the multimodal EEG by combining with the EEG frequency calculation. Second, time-frequency reassignment was used to calculate the reconstructed EEG frequency features and complete the coarse extraction of features. Finally, mutual information was used to extract the frequency features most relevant to the induced EEG, to form a new feature set, and to finally realize the potential frequency multimodal feature extraction. Experimental tests were conducted on the BCI2008 competition dataset, DEAP data set, Impedance data set, as well as the stimulation-evoked EEG signal test data set of an individual in a hospital. The results show that the proposed algorithm is characterized by better EEG preprocessing and signal reconstruction effect, strong subspace clustering of frequency features and less spurious frequencies in the feature sequence; the proposed algorithm has the average accuracy of potential frequency feature extraction up to about 95% and the accuracy of EEG signal classification using the extracted features up to about 97%, MRI and EEG signal fusion effect is good, which further verifies the high efficiency of the proposed algorithm and provides a certain theoretical basis for clinical medical research.