Implementation of Dynamic Neural Network for Neural Decoding Rat Forelimb Trajectory
碩士 === 國立陽明大學 === 生物醫學工程學系 === 104 === Brain Machine Interface (BMI) was a multitudinous research in neuroscience, which has been rapidly developed in recent years. Neural decoding algorithm played an important role in BMIs’ investigates. The neural decoding algorithm mainly decoded ensemble firing...
| Main Authors: | , |
|---|---|
| Other Authors: | |
| Format: | Others |
| Language: | en_US |
| Published: |
2016
|
| Online Access: | http://ndltd.ncl.edu.tw/handle/bhpge7 |
| Summary: | 碩士 === 國立陽明大學 === 生物醫學工程學系 === 104 === Brain Machine Interface (BMI) was a multitudinous research in neuroscience, which has been rapidly developed in recent years. Neural decoding algorithm played an important role in BMIs’ investigates. The neural decoding algorithm mainly decoded ensemble firing patterns into motor commands which could be adopted to drive external devices. Feed-forward neural networks (FNNs) have been developed for BMIs because of their ability of learning nonlinear relationships between neural activity and motor commands. However, they have not outperformed linear decoders substantially due to the lack of network dynamics and difficult understanding of input-output relationship. This study proposed a dynamic neural network (DNN) to predict forelimb movement of a rat according to rat’s neural activity. The rat performed a non-constrained and freely moving forelimb movement in a water-reward lever pressing task. A principal component analysis was adopted to reduce the dimension of inputs and select the features. The DNN had network dynamics which could predict forelimb movement. The proposed DNN was compared to the other neural nework-based decoders. The results showed that the DNN could perform high prediction accuracy in real-time.
|
|---|