Description: Brain Machine Interfaces (BMI), developed initially for individuals with neural disorders, consist of an electrode array that records neural activity, an embedded computer to analyze signals, and a controlled actuator such as a robotic arm. To operate the actuator and ensure the performance of complex actions, sufficient information needs to be obtained from neural recordings. Chronic implants of wireless neural recorders have been used for this task but this type of recordings have shown a new challenge: signals observed by chronic recording electrodes change over time. This ongoing research deals with the problem of designing adaptive real-time neural signal processing algorithms to track the changes of neural signals from chronic recordings in order to maintain good spike sorting accuracy.
In general, spike sorting algorithms involve detecting neural spikes from within background noise, extracting specific features that can represent the signal with the minimum of data points, performing clustering of those features, and then classifying the information of the detected spike into one group of neurons. In order to track changes in the incoming information, improvement to existing algorithms are being made to make them adapt to spike changes, and efficient for hardware implementation and real-time operation.
Author: Sylmarie Davila-Montero