| 摘要: | 通過EEG(Electroencephalography)設備記錄下來的信號，都是大腦神經元共同放電的結果，把這些放電的神經元當作是放電的源頭，而這些放電源會具有不同的方向來放電，經過一定的過程會衰減(皮層、腦脊液、顱骨等組織)，在到達頭皮表面才會被記錄下來，所以這些採集到的信號不能簡單地對應就是在腦袋中放電的位置，所以需要透過溯源的方法來映射腦袋內的情況。
腦電源定位的方法，一般來說有兩種情況，一是正向問題(forward problem)把已知腦電源的位置、大小及方向，正向求解出頭皮電位信號，二是逆向問題(inverse problem)根據已被記錄下來的頭皮腦電信號來反向推估腦內神經元活動的位置、方向及強度大小。在過去的研究中提到，電極偶極子源定位裡用MNE(Minimum Norm Estimate)做逆向求解，會發現和原本放電的腦電源位置會有相對的距離誤差，若透過Depth-weighted去修正，可以縮小電源定位距離誤差，此外，電極的數量和噪聲的多寡和也會對定位準確度造成影響，因此，本研究的目的是提出一個理論方法，以自適性高斯混合模型(Adaptive Gaussian Mixture Model;AGMM)來進行電腦模擬實驗，將先前已知訊號源定位的信息，在inverse時透過MNE，再以Adaptive weighting的方式，整合進低密度腦電圖中(16 channel)的源定位演算法中，來增加EEG的準確度，而本研究結果顯示，當我們再添加權重時，在高密度腦電圖(128channel)中分別在只有訊號源x、y、z方向時，在訊雜比(SNR)給定從1db到15db時，各別的位置誤差是從11.2-3.5mm調降到9.6-2.5mm、20.2-4.9mm調降到18.8-2.7mm以及7.4-1.3mm降至6.9-1.0mm，而在我們訊號源三個方向同時存在時，位置誤差為8.2-1.7mm降至7.0-1mm，而在低密度腦電圖(16channel)上，對訊號源x方向、z方向和三個方向同時存在的情況添加權重後，在我們根據經驗選取上的低密度腦電圖(16channel)，我們都把誤差降至15mm以下，而在我們對於文獻上所選取的低密度腦電圖(16channel)中，原本位置的誤差值範圍有70-90mm、但在經過權重後，我們把誤差值降在55mm以下，因此，在本研究上，我們發現準確度會受到訊雜比(SNR)、電極數量上的多寡以及權重的影響，而權重在對於我們低密度腦電圖上以及當我們給予放電源y方向時，修正誤差的效果是最為顯著的，最後在我們的研究成果上，希望對之後在科研或實際應用上能提供一些幫助。
;The signals recorded by EEG (Electroencephalography) equipment are all the result of the joint discharge of brain neurons. These discharge neurons are regarded as the source of discharge, and these discharge sources will discharge in different directions. After a certain process It will attenuate (cortex, cerebrospinal fluid, skull and other tissues), and it will be recorded when it reaches the surface of the scalp, so these collected signals cannot simply correspond to the location of the discharge in the brain, so it is necessary to map the brain through the method of traceability case.
Generally speaking, there are two methods for the location of the brain power supply. One is the forward problem (forward problem), which is to solve the known brain power position, size and direction forward to obtain the scalp potential signal, and the other is the inverse problem (inverse problem). Based on the recorded scalp EEG signals, the position, direction and intensity of neuron activity in the brain can be reversely estimated. As mentioned in the past research, using MNE (Minimum Norm Estimate) for the reverse solution in the location of the electrode dipole source, it will be found that there will be a relative distance error from the original discharge brain power source position. If it is corrected through Depth-weighted, it can be In addition, the number of electrodes and the amount of noise will also affect the positioning accuracy. Therefore, the purpose of this study is to propose a theoretical method to use the Adaptive Gaussian Mixture Model (Adaptive Gaussian Mixture Model; AGMM ) to carry out computer simulation experiments, the previously known signal source location information is integrated into the source location algorithm of the low-density EEG (16 channel) through the MNE in the inverse, and then in the form of Adaptive weighting. Increase the accuracy of EEG, and the results of this study show that when we add weights, in the high-density EEG (128channel) when there are only signal source x, y, z directions, the signal-to-noise ratio (SNR) gives When it is set from 1db to 15db, the respective position errors are reduced from 11.2-3.5mm to 9.6-2.5mm, 20.2-4.9mm to 18.8-2.7mm and 7.4-1.3mm to 6.9-1.0mm, while When the three directions of our signal source exist at the same time, the position error is 8.2-1.7mm and reduced to 7.0-1mm. On the low-density EEG (16channel), the signal source x direction, z direction and three directions exist at the same time After adding weights to the situation, in the low-density EEG (16channel) we selected based on experience, we all reduced the error to less than 15mm, and in the low-density EEG (16channel) we selected for the literature, The error range of the original position is 70-90mm, but after weighting, we reduce the error value below 55mm. Therefore, in this research, we found that the accuracy will be affected by the signal-to-noise ratio (SNR) and the number of electrodes. The influence of the amount and the weight, and the weight is the most significant effect of correcting the error on our low-density EEG and when we give the y direction of the discharge source. Finally, in our research results, we hope that it will be useful in scientific research or in the future. It can provide some help in practical application. |
