為了研究人類最後的黑盒子“大腦”，相關的研究工具相繼推出以幫助人們解答對大腦的疑問，當中以功能性核磁共振儀和腦電波儀的應用最為廣泛，功能性核磁共振影像具有良好的空間解析度，不僅應用到醫療上幫助我們了解大腦的組織，更提供我們以認知實驗方式來對大腦功能的定位，而腦電波圖的良好時間解析度，也幫助我們了解大腦真正的活動－神經元反應，及大腦處理認知歷程的時間性，但由於這些工具所得到信號都是極為微弱且特性複雜，使得我們在處理這些訊號上需要一個良好的方法來幫助得正確且精準的大腦反應。 為了偵測功能性核磁共振影像的活化區域和腦電波的來源定位，我們提出一個視覺化工具(V map)，藉著視覺的檢查視覺化工具所產生的影像，使用者能很容易定義出功能性核磁共振影像的活化區域和腦電波的來源位置，所提出的視覺化工具完全利用活化像素的空間連結特性，這是一個重要的因素在決定大腦的活動，這個方法的優點是不需要一個先備的知識，我們使用一些資料集來証明這個方法的應用性，並利用ROC（receiver operating characteristic）分析，來探討視覺化工具及其他方法運用在功能性影像分析的效能特性，以及在偵測未知之功能性反應的精確性。結果顯示，在適當的信號變化對雜訊比（contrast-to-noise ratio）之下，視覺化工具確實能有效地偵測出功能性核磁共振影像活化區域和腦電波的來源。 在功能性核磁共振影像(fMRI)中，我們藉由模擬信號和實際人體實驗的資料來証實V map的強健性和有效，比較到以模式為主的方法，V map能有效的偵測fMRI的活化區域而不需要一個預期血液動力反應的先備知識，比較到資料導向為主的方法，V map提供幾個有魅力的特色：1)處理時間（含分群演算法）不需要很長；2)決定最佳的分群數量已不再是問題；3)資料的順序和群聚中心的隨機初始不再是問題；4)使用者不需要去確定真正符合活化區域的群聚。 在腦電波(EEG)中，我們也利用模擬信號和實際人體實驗的資料來証實V map的堅實和有效，比較到偶極子為主的方法，V map能有效的偵測EEG的來源位置而不需要假設任何的偶極子來源數量，比較到無母數統計方法(statistical nonparametric mapping method)，V map提供幾個有魅力的特色：1)處理時間不需要很長；2)單一樣本資料就可檢測來源定位；3)樣本數不會影響結果的信賴度；4) V map對雜訊的影響度低。 另外，V map有能力偵測區域中像素具有類似活化的參數，但可能不同於一個真正模式函數，假如這些未預期的活化區域能更進一步檢閱，那麼會有更多資訊被顯露出來。 For opening the black box of human, understanding the brain, many tools have been applied to help people probe the questions of brain. Functional magnetic resonance imaging (fMRI) and Electroencephalography (EEG) are of the most extensive application. fMRI has very high spatial resolution which not only helps us understand tissue of cortex, but also provides us information of the functional sites of brain in cognitive experiment. On the other hand, EEG has very high temporal resolution; it helps us to enable us to see the direct action of brain (neurons) and the sequential processing of brain on cognitive operation. However, the neural signals are very weak and usually present complex property in these tools. We need a better method to get correct and precise brain action on processing singles. We propose a visualization map for the detection of activated regions in fMRI and of the located source in EEG. Via the visual examination of the map, the user can easily identify activated regions in the fMRI and the located source in EEG. The proposed visualization map fully utilizes the property of the spatial connectivity of activated pixels which is a key factor on determining the significance of activated regions in cortex. The advantage of this technique is people can identify regions of activation without any priori knowledge. Several datasets have also been employed to verify the broad applicability of the technique. Receiver operating characteristic (ROC) analysis has been used to evaluate the performance under many conditions. Results show that with certain contrast-to-noise ratios, visualization map can detect the functional activation in fMRI and source location in EEG. In fMRI, the robustness and effectiveness of the V map are demonstrated by the simulation results on an artificial data set and a real life one. Comparing to model-based methods, our V map can effectively detect regions of activation in fMRI without a priori knowledge of expected hemodynamic responses. Comparing with data-driven-based methods, the V map offers several appealing properties: 1) the processing time involved with clustering algorithms is no longer needed; 2) the determination of the optimal cluster number is not necessary; 3) the order of data and the random initialization of the cluster centers are not necessary; 4) the user no longer needs to determine which cluster really corresponds to the activated regions. In EEG, we also use an artificial data set and a real life one to demonstrate the robustness and effectiveness of the V map. Comparing to dipole-model method, V map can efficiently detect location of sources in EEG without a priori assumption on the number of dipoles in the brain. Comparing with statistical nonparametric mapping method, V map offers several appealing properties: 1) the processing time is no longer needed; 2) one condition is sufficient to define sources location; 3) the number of samples does not influence the reliability of result; 4) correcting ratio of V map is very low in the different SNR. In addition, the V map is capable of detecting regions within which pixels have similar activation patterns which may be different to the real model function. If these un-expected activation regions can be further inspected more information may be revealed.
