以薄片逆回歸對大鼠運動皮質之神經元動作電位編碼進行前肢運動軌跡預測

Abstract

在神經科學的研究領域中，其中一項重要的研究是腦機介面的發展。而在腦機介面中，運動皮質區的神經群動作電位之解碼演算法更是扮演著不可或缺的角色，它是用來解析運動皮質區的神經群放電並且預測出生物體的運動軌跡進而去控制外在的神經義肢，而目前在這個領域中已經有許多的演算法被開發出來，甚至運用在實體上。然而在這類的研究中，隨著電極的數量越來越多，資料的維度也越來越大。因此，我們想要開發出一個保有原來資料特性，卻可以使資料的維度大大降低的方法。 在本研究裡，我們設計了一個老鼠實驗去驗證我們所開發的演算法，薄片逆回歸。薄片逆回歸利用腦部的神經訊號以及從影像中所擷取的軌跡來建立模型，並且跟目前已經存在這領域的兩個演算法，群體向量法(Population Vector Algorithm) 和最佳線性估測法(Optimal Linear Estimation)做比較。而在軌跡預測結果的方均根誤差中，我們發現薄片逆回歸的方法比群體向量法低了28.26%、比最佳現性估計法低了35.19%，並且在軌跡預測的相關係數中達到(0.8729,0.8647) 的高相關度，因此證明了我們所運用的新方法比另外兩個線性的方法還要好。除了演算法開發之外，我們也發現老鼠在做壓水動作的期間，場電位在12赫茲以下出現了能量上升的情形，間接證明了我們所收到的神經訊號是與此壓桿行為是有關係的，因此得知我們用此訊號來開發演算法是合理的。

In the neuroscience field, one of the key researches is Brain Machine Interfaces (BMIs). Neural decoding algorithm development plays an important role in BMIs’ investigates. Decoding algorithm in primary motor cortex is able to achieve the goal about controlling the external computer cursors or prosthetic limbs. Many algorithms have been developed in current, and population vector algorithm has been implemented in real time among of linear decoding. The data dimension in this field has become more and more plenty with the implanted electrode arrays increasing. Therefore, we want to develop a decoding algorithm is capable of extracting intrinsic information and reducing the data dimension. In our research, we have designed a depressed experiment for rat to demonstrate our developed algorithm, Sliced Inverse Regression (SIR), which is used to build a model between neural activities from brain and forelimb’s trajectory from video. Moreover, compare the prediction with two current linear decoding algorithms, Population Vector Algorithm (PVA) and Optimal Linear Estimation (OLE). In the root-mean square error criterion (RMSE), SIR has lower RMSE in 28.26% than PVA, and lower in 35.19% than OLE. Additionally, SIR achieved a high correlation (0.8729, 0.8647) at trajectory prediction. Therefore, we can know that SIR has a better prediction than other in our case. In addition to algorithm development, the specific pattern which the power had an obvious increasing under the 7~12 Hz for the power spectrogram of local field potential was occurred during behavior task. Therefore, indirectly prove the relationship between neural activities and behavior task so that the neural activities were significant to develop decoding algorithm.