Loading...

Neural Spikes Sorting and Decoding to Task Extraction

Samiee, Soheila | 2011

1141 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 42138 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Shamsollahi, Mohammad Bagher
  7. Abstract:
  8. Brain is the most complicated organ of body which controls the activity of all other organs. Understanding its function and its language could give us a direct communication pathway for connecting injured motor organ and it could be useful for functional repairing. Neurons are atoms of a vast network that generate the brain signals. Processing these signals would help to translate brain’s language and has three main stages: spike detection from signal, spike sorting, and intention extraction from encoded signal. In this research, we use a dataset of rat’s extracellular recordings during a time interval in which a rat pressed the liver several times to receive water as an award. Scince spikes were detected by the recording setup, before any post processing;we improve this stage’s results. Next step is spike sorting which uses a novel idea based on Hidden Markov Models (HMM). This method leads to a desirable result with normalized MSE of 0.34 and the under ROC curve area of 0.96. In next step, we used the result of former steps to extract a task with decoding the train of spikes. Initially, we exploit an effective neuron selection algorithm and also compensate the delay of each neuron separately. Then, two task extraction methods are investigated. One is using Radial Basis Function Network to estimate the task and the other is extraction using the innovation signal of Kalman filter. These methods are successful in tracking the hand and finding the status of the liver pressure in such a nonstationary continuous problem. The area under ROC curve for these two methods are 0.955 and 0.962 respectively which shows better performance than the result of two other common methods of using Optimum Linear Estimation and Kalman filter in extracting the task whose under ROC curve area are 0.816 and 0.838 respectively.

  9. Keywords:
  10. Hidden Markov Model ; Extracellular Recording ; Neural Spikes ; Spike Sorting ; Task Extraction

 Digital Object List