在本計畫中，我們提出了透過同步腦電位與血氧濃度變化監控程序來估計癲癇波發作在大腦中的源頭位置，並以透過顱內腦電位等腦內紀錄的結果，對於估算的溯源結果，進行評估，以確認其臨床上的可行性與可用性。本子計畫擬利用深度學習的方式，分別對同時擷取的腦波與功能性磁振造影資料進行分析，針對臨床醫師所指出的癲癇波病兆與非病兆的訊號區段，在多種神經影像資料中，利用深度學習中循環神經網路的長短程記憶模型，進行分類的處理，嘗試找出各種不同神經影像之間，在癲癇波發生的前後，具有何種對應的關係。這樣的對應關係，將可以進一步評估非侵入式神經影像方式作為定位癲癇發作位置的可行性，從而作為日後臨床上評估癲癇發作的重要工具，協助癲癇手術術前評估的電極定位規畫。而循環神經網路網路的使用，將協助評估癲癇波發作前是否有某些腦區的活動隨著時間有能量累積的作用，這些具有能量累積作用的腦區，或可協助定位具有多源的癲癇發作。同時本子計畫也規畫將這個循環神經網路的算則應用於其他不同認知作業所獲得的神經影像資料，用以評估本方法在其他認知作業神經影像資料分析上的可行性。 ;In this project, we propose to estimate the sources of epileptic activity using simultaneous EEG and fMRI examination with the combination of internal recordings, such as stereotaxic EEG (sEEG) as used in the clinical practice for confirmation. To analyze the multi-modal neuroimaging data, we propose to use the deep learning technique of recurrent neural network (RNN), especially the long short-term memory (LSTM) model, to find the brain areas in which the temporal dynamics are highly associated with the epileptic spikes as labeled by the physicians in the EEG data. In so doing, we expect to find the linkage between different neuroimaging measurements commonly associated with the same epileptic spike event. Such an association may be indicative of the feasibility of using non-invasive neuroimaging modalities for estimating the epileptic focus (or foci). Also, the use of RNN provides the possibility to evaluate if any energy accumulation in specific brain areas that are further associated with the seizure onset. We hypothesize that such an energy accumulation phenomenon may be informative for indicating the spreading of a seizure onset. Furthermore, we are also to expand the RNN deep learning method to the functional neuroimaging data obtained from the experiment of a regular cognitive task performance to test the applicability of the proposed method for other types of cognitive neuroimaging data. Here, we will try to analyze the fMRI data from a simple Go/NoGo task performance using the proposed and devised analytic method in this proposal.
