Independent Component Analysis In Medical Signal Processing Applications

In the field of medical signals processing, the signals are corrupted by all kinds of artifacts, such as the power line influence, Electro-Oculogram (EOG), Electrocardiograph (ECG). All these artifacts are from mutually independent sources. How to extract the useful components from the corrupted raw data still remains a challenging problem.Independent Component Analysis (ICA) emerged as a novel technique for blind source separation. This method is based on high order statistics. ICA is to separate the original signals into several independent components by selecting a particular criterion and an optimal algorithm. Though the specific information of the sources and mixing system are unknown, the probability for the source decomposition can be realized by using the principle of statistical independence. ICA can not only remove the correlation between signals, but also make use of the higher order statistics. Therefore, ICA has a better performance than Principal Component Analysis (PCA) and is widely used in many fields.The definition and principles of ICA was presented firstly in this paper. It mainly focuses on the application of ICA in the filed of biomedical signal processing. The results of this study can contribute to the clinical application and the cognitive sciences and brain science. In this article we adopt the algorithm of Fast ICA and Infomax ICA for image separation and EEG artifact reduction in the Matlab environment.In this paper, the ICA of noisy signals is discussed. In the simulation, the technique of wavelet threshold de-noising and the algorithm of Fast ICA are both studied with computer simulation of noisy image separation. The simulation results show that for the mixed images with additive white Gaussian noise, it's better to de-noise the images before applying ICA than to apply ICA first and then de-noisethe independent components.We also combine the AR model with the ICA model to analyze and classify the mental EEG with BP neural network. We first purify the raw mental EEG signals using ICA, and the artifacts and noises of EOG and power line interface are removed. After ICA is applied to extract the ICs of the cleaned EEG signals, the AR model is used to calculate the AR coefficients. Finally, the BP neural network was trained to classify the signals. The testing simulation shows that the method can get a recognition rate of 80 - 90%.