Transient brain power supply positioning method and system based on non-negative block sparse Bayesian learning

Abstract

The invention discloses a transient brain power supply positioning method and system based on non-negative block sparse Bayesian learning. The method comprises the steps: firstly collecting a transient multi-channel EEG, calculating a sample covariance matrix, and weighting each column of the matrix, which can be represented as non-negative block sparse representation on each brain region after Laplace smoothing; then setting an iteration stop condition and an initial value of a non-negative brain region power sparse support vector; iteratively updating a posterior mean value and a covariance of the non-negative brain region power vector based on non-negative Gaussian distribution, and updating a non-negative brain region power sparse support vector according to the posterior mean value and the covariance; and finally, giving a source positioning result by using the latest non-negative brain region power sparse support vector. According to the invention, brain region non-negative block sparse representation of covariance vectors is utilized, expectation and variance of non-negative Gaussian posterior distribution are combined, and the transient EEG source localization NNBSBL method is given; and the method does not need to predict or estimate a noise covariance matrix, is high in positioning precision and resolution, and provides a technical means for the fields of cognitive psychology, brain-computer interfaces, nerve diagnosis and treatment and the like.

Description

technical field [0001] The invention relates to the technical field of electroencephalogram (EEG) traceability, in particular to a method and system for locating transient brain power sources based on Non-Negative Block Sparse Bayesian Learning (NNBSBL). Background technique [0002] Electroencephalogram (EEG) is a potential signal that can be recorded along with brain activity. It is of great significance for studying brain activity, evaluating brain function, clinical and psychological research, and brain-computer interface. EEG includes two forms: transient EEG and steady-state EEG. Transient EEG is manifested as a short-duration EEG waveform with certain characteristics, while steady-state EEG is associated with a specific frequency for a longer period of time. Transient EEG can be induced, such as various forms of transient evoked potentials, or spontaneously, such as the spike (sharp) slow complex during epileptic seizures. [0003] The source location problem of EEG,...

AI Technical Summary

Problems solved by technology

[0006] Invention patent application (brain source location method based on sparsity constraints, application number 200610021584.6) integrates Lp norm regularization constraints into the FOCUSS algorithm, thus giving an EEG source location method, which does not consider current dipoles The source is block-sparse features on the cortex, and the resolution and estimation accuracy of the Lp-FOCUSS algorithm are low

[0007] Invention patent application (a brain power location method based on Granger causality, application number 201410636555.5) by analyzing the Granger causality between each electrode channel and other electrode channels for source location, in order to solve the EEG inverse problem However, the location of the source determined by this method is in the lower cortex of the scalp electrode, and the mapping effect of the lead field matrix is ​​not introduced, so the accuracy of source location will be affected

[0008] Invention patent application (an LSTM-based EEG signal source location method, application number 201910178711.0) trains an LSTM neural network and inputs multi-channel EEG data into the neural network to directly obtain the location of the source. Neither the training nor the testing of the neural network takes into account the sparse distribution of current dipole sources on the cortex, and outputting binary results can easily lead to high false positives or false negatives in EEG source localization results.

[0009] Invention patent application (Sparse Bayesian learning-based steady-state evoked response brain source localization method, application number 202010001431.5) provides a SBL source localization method for steady-state EEG, but this method is for steady-state EEG rather than Transient EEG, and still not accounting for block sparse distribution of current dipole sources over the cortex

[0010] Invention patent application (brain power location method, system and equipment based on population diversity control, application number 202011096444.1), proposes a quantum particle swarm algorithm based on population diversity control, and improves the global traversal of the algorithm by controlling the population diversity of the algorithm This method still belongs to the category of dipole matching method, and it needs to preset the number of current dipole sources, which limits its practicability

[0013] 1. Insufficient consideration of the block-sparse distribution a priori of current dipole sources on the cortex

[0014] 2. Need preset noise power

[0015] 3. Need to preset the number of current dipole sources

[0016] 4. It cannot solve the problem of spatially non-uniform noise, and simultaneously estimating the noise power in the algorithm will affect the positioning accuracy of the current dipole source

[0017] 5. The positioning target is often the time-varying signal itself rather than the power of the signal on each vertex, and the non-negativity of the signal power is not used to improve the effect of EEG source positioning

Images