• Top page
  • Timetable
  • Per session
  • Per presentation
  • How to
  • Meeting Planner



Attention and Perceptual Integration

開催日 2014/9/11
時間 11:00 - 12:00
会場 Poster / Exhibition(Event Hall B)

Support Vector Machineに基づくラット海馬神経活動を用いたレバー押し予測法
A support vector machine based approach for lever pressing prediction using rats hippocampal activity

  • P1-243
  • 田中 徳文 / Norifumi Tanaka:1,2 青西 亨 / Toru Aonishi:2 Genci Capi / Capi Genci:3 臼井 弘児 / Kouji Usui:1,4 川原 茂敬 / Shigenori Kawahara:1,4 
  • 1:富山大院生命融合科学教育 / Grad. Sch. Innovative Life Science, Univ. Toyama 2:東京工大院 総合理工知能システム科学 / Interdisciplinary Grad. Sch. Science and Engineering, Tokyo Inst. Technology 3:富山大工 / Dpt. Electrical and Electronic Systems Engineering, Fac. Engineering, Univ. Toyama 4:富山大工生命工学 / Dpt. Life Sciences and Bioengineering, Fac, Engineering, Univ. Toyama 

Most of the BMI (Brain Machine Interface) studies have been focused on developing techniques to control machines using a set of neuronal activities in the motor cortex, the final motor output area. However, information from other brain areas could also contribute to an increase of the percentage of the correct action prediction by providing an information on the brain state to prepare the motor outputs. In this study, we analyze hippocampal LFP (Local Field Potential) during lever-pressing task in rats. Three male Wistar/ST rats were first trained to press the right or left lever under a body-restrained condition to control the food pellet carried by e-Puck mobile robot to get near the mouth. After reaching a sufficient skill level, an electrode of twisted wires for LFP recording was implanted in the hippocampus. The LFP data during the lever-pressing task were analyzed off-line using SVM (Support Vector Machine) and radial based function were utilized to predict the lever pressing. For optimization of SVM parameters, we used a set of LFP power spectra with a time window ranging from 1 to 9 sec around the time of the lever-pressing and 3-sec time window outside the period of lever-pressing. Then, we evaluated the difference in distance among various power spectra using 10-fold cross validation. We found an elevation of task-related theta oscillation 2 to 3 sec before lever pressing, irrespective of the right or left lever. The results show that radial basis function achieved a 70 % to 90 % correct lever pressing prediction rate using the LFP data. The optimal time window of power spectrum was 1 to 3 sec. In addition, the results show that SVM and the radial basis function applied to the hippocampal LFP improve the performance of BMI systems.

Copyright © Neuroscience2014. All Right Reserved.