人類的大腦是目前所見最複雜的器官亦是掌管我們生理與心理活動的中樞，為了探討腦動力學以及形形色色的認知功能，從19世紀末葉以來相關的研究持續地發展中。其中的一個領域其研究工作是分析腦神經活動產生的電磁訊號來對其作定位。近年來由梁偉光和王敏生提出的方法”SIMN”能對稀疏的腦磁源產生的無雜訊訊號做完全精確的定位，然而其定位的精確度因雜訊的干擾而下降許多，故我們想提出一個方法來改善SIMN對於有雜訊時的結果。其主要的構想是把訊號向量投影在lead field的空間裡來選擇ㄧ些格點（此為腦神經可能活動的位置），並應用SIMN來對這些格點做進ㄧ步的定位，然後再用投影的方式選出新的格點和用SIMN來定位。經過一連串重複的步驟直到選出的格點穩定為止。在程式中，我們使用真實的頭模型並用數值理論的方式產生欲分析的腦磁波訊號。我們針對1-3個源的無雜訊干擾以及有雜訊時的情況分別做了模擬，並與其他方法譬如WMN，sLORETA，FOCUSS，SSLOFO，SIMN的結果做比較。對於無雜訊的情形，這個投影的方法能完全定位源的分佈。另外對於有雜訊的情況，雖然此方法的定位命中率仍受雜訊的干擾而降低，不過的確改善了SIMN對於雜訊時的結果，而且其定位的精準度是所有比較的方法裡最高的。 The human brain is the most complex organ we have ever seen and it controls our physical and psychological behaviors. Since late nineteenth century, remarkable advances have been made in exploring the dynamics and various functions of human brain. One area of research is to localize the neuronal activities within the brain by analyzing the electromagnetic recording from them. Recently, an approach referred to as “SIMN’ proposed by W. K. Liang and M. S. Wang can obtain exact inverse solutions for noiseless MEG signals from sparse point sources provided that there is no significant cancellation of signals among the sources. However, the localization power of SIMN decreases with noises. Here we would like to propose a method that could yield better localization power than SIMN for noisy signals. The main idea of our inverse method is to select a set of grid points as possible positions of sources by projecting the data vector to the lead field space. The algorithm also applies SIMN to determine the source positions. In this recursive procedure, we select new grid points by projecting the data vector to the lead field space and applying SIMN repeatedly until the set of selected grid points stabilizes. The realistic head model is used and MEG data are generated theoretically. Both noise-free and noisy MEG data for one-, two- and three-point-source cases are studied in the simulations. We demonstrate the method and compare its performance with WMN, sLORETA, FOCUSS, SSLOFO and SIMN. For noiseless cases, exact inverse solutions for noiseless MEG signals from sparse point sources are obtained by our method. For noisy cases, although the localization ability of our inverse method still decreases with noises, the resolution power of our method is better than that of SIMN. The results show that our inverse method has the highest localization power among these methods.
