Cooperative Neural Decoding of Goal-directed Forelimb Movement with Internal Error Feedback Scheme from Rodent Motor Cortex
Presentation Number:0013 Time:10:55 - 11:07
|
Chia-Jung Yeh, Yi-Ting Chou, Shih-Hung Yang, Yu-Chun Lo, Yu-Hao Lan, Chen-Yang Hsu, Kuan-Yu Chen and You-Yin Chen
In Brain Machine Interfaces (BMI), goal-directed movement coordinated the reaching target in external environment through vision, and then the upper limb would be drove to execute the motor movement. To accomplish the goal success-fully, the reaching movement required on-line control to modi-fy or update the ongoing hand movement through visual feed-back in mostly condition. Moreover, during the on-line correc-tion of goal-directed movement, the visually derived relative position of hand and target was the key information for modi-fying the movement state. It had been found that the neural population in premotor cortex of monkeys encoded this infor-mation: visually derived relative position of hand and target. In aspect of intracortical micro stimulation, the forelimb areas in secondary motor cortex (M2) of rodent was considered equivalent to premotor area in monkeys. In this study, we demonstrated that the target-hand relative position could be decoded from the neural activity in M2 of rodents successfully. This decoding model worked as an internal error feedback scheme to decode the visual-feedback-related information from M2. By combing this internal error feedback model to present commonly used neural decoder in BMI could significantly improve the decoding performance.
|
|
An Embedded Real-time Neural Decoding System for Prediction of the Rat Forelimb Movement
Presentation Number:0159 Time:11:07 - 11:19
|
Yi-Hsin Yeh, Bo-Wei Chen, Chia-Jung Yeh, Yi-Ting Chou, Yu-Chun Lo and You-Yin Chen
Brain Machine Interface (BMI) was the method of transform-ing mental thoughts and imagination into actions. A real-time BMI system improved the quality of life of patients with se-vere neuromuscular disorders by enabling them to communi-cate with the outside world. At present, the important aspect of the real-time implementation of neural decoding algorithms on embedded systems has been often overlooked, notwith-standing the impact that limited hardware resources have on the efficiency/effectiveness of any given algorithm. In this study, the prediction model was built based on the neural activities (spikes) from rodent primary motor cortex (M1) and their corresponding trajectory of rodent forelimb by decoding algorithms, kernel sliced inverse regression (kSIR). Mean-while, a cost effective way of implementing and designing a demonstration platform for BMI research was presented, featuring a low-cost hardware implementation based on an open-source electronics platform Raspberry Pi. Here, the nonlinear decoding algorithm was implemented as separate signal analysis modules for the real-time decode of end effec-tor trajectory.
|
|
|
|
Robust Decoding Animal Forelimb Movement Using Kernel Sliced Inverse Regression with Multimodal Neural Signal
Presentation Number:0351 Time:11:43 - 11:55
|
Chin Chou, Shin-Hung Yang, Yu-Chieh Lin, Guan-Yu Chen, Yi-We Lee, Yu-Chuan Hung, Chun-Hang Hsu, Yu-Chun Lo and You-Yin Chen
Accurate long-term control of prosthesis has been a significant issue for neural signals recorded at brain machine interface (BMI). The accuracy, stability and longevity are the key considerations of BMI applications. However, a lever-pressing task and electrode recordings were used to investigate the rodents encoding of hand velocity and trajectory in primary motor cortex (M1). The multiscale neural signals decoded by brain machine interfaces algorithm can be divided into two types: action potentials (spike) from individual neurons and local field potentials (LFP) from extracellular space around neurons. Since the different superiority of spikes and LFPs, this study proposed a decoding algorithm- kernel sliced inverse regression (kSIR) which combined spikes and LFPs as the multimodal inputs to decode the contents of the mind with high accuracy and stability. Results showed that the stability and accuracy were significantly improved, where the accuracy was improved from 0.88 ± 0.059 to 0.93 ± 0.061 for X-axis and 0.90 ± 0.022 to 0.97 ± 0.024 (R-squared, Mean ± SEM) for Y-axis. This implementation favorably lends itself toward the long-term decoding approach which can provide accurate real-time decoding of neural signals over periods of days.
|
|