Machine Learning of Functional Magnetic Resonance Imaging Data for Brain Decoding

• 作 者: NorikiKoike;YutakaHatakeyama;ShinichYoshida;
• 机 构: Kochi University of Technology,Japan
• 出 处: 第三届先进计算智能与智能信息学国际研讨会(The Third International Workshop on Adv
• 发表日期: 2013年期

　　Brain decoding is a technology that reads the a humans internal state or behavior from his/her brain activity obtained from functional magnetic imaging or other devices measuring brain activity.Core technology of brain decoding is a decoder,which is a machine learning algorithm such as support vector machine.Improvement of machine learning precision and speed of convergence is important issues and then the purpose of this research is comparison of machine learning algorithms for the brain decoding application.In this study,we construct the brain decoder using four machine learning algorithms,that are support vector machines,neural networks,random forests,sparse logistic regression.We conduct a brain decoding experiment of the game of “paper,” “scissors,” and “stones.” The decoder decodes a subjects hand from fMRI measurement.The result shows that the highest performance in both accuracy and speed are achieved by SVM using the linear kernel,and those of the lowest are achieved by neural network.