| 摘要: | 本研究使用動態因果模型估測出大腦網路結構和控制大腦連結的參數，目的是將腦電波資訊使用動態因果模型之誘發響應研究中風病患復健前後對於大腦運動網路結構以及連結頻率上的動態改變。此外，我們依照不同的復健計畫將病人分成兩組，以傳統復健的受試者納入control組，以玩虛擬實境(Virtual-Reality,簡稱VR)復健遊戲治療的受試者納入study組。
本研究共收錄了21受試者，11位study組的病患(平均年齡: 54.73±10.93歲)和10位control組的病患(64.4±8.72歲)。每位受試者經歷了24小時的復健訓練，一週復健四天，一天復健一小時，其中復健時study組玩一小時類似傳統復健的虛擬實境遊戲，而control組則是執行一小時的傳統復健。在治療前和治療後(完成復健訓練)每位受試者會接受實驗記錄30頻道腦電波，取樣頻率為2000Hz。在實驗中，我們要求所有受試者執行80次患側上肢的動作，例如肩膀或手肘作曲屈/伸展的動作(動作是由治療師判斷受試者狀況而決定的)。腦電波訊號前處理先抑制60Hz的電線干擾的影響，再將訊號以治療師的口令(受試者每次開始執行上肢動作都是聽從治療師的口令)為中心裁切出 -800到 +800毫秒的資訊，然後再濾波使用4Hz和48Hz的帶通濾波器，接著將腦電波資訊以動態因果模型之誘發響應分析估測大腦網路模型之結構(本研究使用的網路模型有六種，其中都是由雙側的初級運動皮質區和前運動區以及輔助運動區所構成)，估測方法是由貝氏分類決策理論選擇出最能代表個人和群體的大腦網路結構模型，
得到選擇出最適合代表大腦網路的模型後將模型內資訊執行多變異數分析、成對樣本t檢定分析、兩個樣本t檢定分析。至於臨床上我們則透過治療師使用三種評估工具(FM-UE、TEMPA、WMFT)來評估受試者之上肢運動功能。
實驗結果在受試者上肢運動功能的評量分數上，兩組受試者在經由復健治療後都有顯著的進步(p<0.05)，而且在兩組之間並沒有顯著的差別，這意味著VR復健之成效可媲美傳統復健之成效，另外依照貝氏分類決策理論選擇受試者執行上肢動作任務時最能代表的大腦網路，結果不論在前後測時均選擇了擁有對側優勢的模型來代表兩組受試者之大腦網路，這意味著大腦在經由不同的的復健方式後所產生功能性的恢復之策略是相同的，此外，我們將兩組所選擇之模型內的連結強度進行統計分析，在同組內前後測比較中觀察到control組在復健後同側大腦內的連結強度減少，study組在復健後則是利用beta頻帶溝通的連結強度減少，最後我們比較了兩組執行了不同復健計畫後在大腦網路連結改變之差異，我們觀察到兩組相比顯示control組的改變優於study組的改變有四條連結，而study組的改變優於control組的改變卻只有兩條連結。
最後，本研究實驗中不只使用動態因果模型估測大腦執行運動任務時的網路結構，還進一步探討大腦網路結構內彼此連結相關之差異，這與先前的EEG研究和fMRI研究相比，我們更能觀察到區域彼此間連結的改變，呈現一個更完整更合理的分析結果，我們也的確觀察到VR復健和傳統復健對於大腦運動網路的影響；在未來，希望可以將這些資訊加以證明VR復健之成效優於傳統復健以及應用在中風後評估患者之臨床研究上的參考。; In this study, we aimed to investigate the cerebral re-organization of motor networks in response to rehabilitation after stroke using Dynamic Causal Modelling for induced responses (DCM_IR) as measured with electroencephalography (EEG). Specifically, the difference in changes due to Virtual-Reality game based rehabilitation (VR) and conventional rehabilitation (CR), including the architecture of the motor network and the coupling parameters govern it resulting from DCM, was examined and compared.
21 stroke patients were recruited in this study and divided randomly into two groups: 11 for VR group (average age:54.73±10.93) and 10 for CR group (average age:64.4 ±8.72). All subjects underwent a totall 24 hours training program with the frequency of 1 hour a day, four days a week. The daily treatment for the VR group consisted of playing 1 hour of home-made VR game which was specially designed to match the conventional CR protocol while the CR group underwent 1 hour route CR treatment. 30 channel EEG were acquired with 2000 Hz sampling rate before (pre-treatment) and after (post-treatment) completed the rehabilitation training. During the EEG acquisition, all subjects were asked to perform about 80 trials of simple upper-limb movement, like shoulder or elbow flexion-extension (subjective to the subject' ability and decided by the therapist), using their affected hand. The EEG data were filtered with 60Hz stop-pass filter to remove power line noise and epoched form -800 to +800 ms where the time zero indicated the inhibition of movement as commanded by the therapist. The epoched data were further filtered with 4~48 Hz band-pass filter and the entered DCM_IR for network identification. Six plausible models, comprising bilateral primary motor cortex (M1), premotor cortex(PM)and (supplementary motor area)SMA, were tested and Bayesian model selection (BMS) was used to selected the model that can explain the data best at both the single subject and group level. Statistic test on the estimates of the best model were performed using analysis of variance (ANOVA), paired t-test, two sample t-test. Clinical measures included FM-UE, TEMPA, WMFT were also conducted by the therapist pre- and post-treatment.
The statistic test on FM, TEMPA, WOLF shows there has a significant difference between pre- and post-treatment but no significant difference between VR and CR groups, indicating that VR based rehabilitation has the similar treatment effect as the convention CR does. The BMS identifies the model with contralesional M1 dominating the network structure for both groups and for pre- and post-rehab, suggesting that the strategy used by the brain for functional restoration is identical: to be as normal as before, irrelavent to the means of intervention. We compared the changes in coupling parameters and found that CR group is better than VR group with four connections and VR group is better than CR group with two connections only.
In conclution, we investigated the architecture of the motor network during task by DCM and the difference of connection between architecture of the brain. We found a more complete result by focus on connection changed from cortical area compared with previous studies and we observed that cerebral motor networks indeed affected by occupational therapy and VR based rehabilitation. We believe the outcome of this study could be an evidence that VR based rehabilitation is better for stroke patient and benefit the study of motor recovery during rehabilitation n after stroke in the future. |
