Predicting stimuli performed using artificial neural network | |
---|---|
( Volume 2 Issue 10,October 2016 ) OPEN ACCESS | |
Author(s): | |
Rafael do Espirito Santo | |
Abstract: | |
This paper presents a new method of inferring mental task performed by subjects during the completion of a block designed functional magnetic resonance imaging (fMRI). The proposed method uses Principal Components Analysis (PCA) formulation and a Multilayer Perceptron (MLP) classifier. The inference is performed based on images derived from paradigms made by subjects during an fMRI experiment. Using these images, distinct activation maps are generated by XBAM software for visual, auditory, and hands movements paradigms. On individuals basis XBAM detects a multitude of brain areas in each paradigm with great variability. The most frequent are: left precentral gyrus (in 95% of the cases) and superior right cerebellum (87%) during the right hand movement; right precentral gyrus (88%) during the left hand movement; right (93%) and left (91%) middle temporal gyrus for the auditory paradigm; right (90%) and left (88%) lingual gyri during visual stimulus. The maps with detected areas are used to train the MLP network in classifying corresponding paradigms. The MLP is trained in a reduced-dimension feature space, obtained through PCA of original feature space. In order to demonstrate the viability of the proposed method, inferences of paradigm performed by 54 healthy subjects is presented. The paper also presents the influence of the number of Principal Components (PC) on the performance of the MLP classifier which in this work is evaluated in terms of Sensitivity and Specificity, Prediction Accuracy and the area Az under the receiver operating characteristics (ROC) curve. From the ROC analysis, values of Az up to 1 are obtained with 60 PCs in discriminating the visual paradigm from the auditory paradigm. Due to the great amount of areas detected in each stimulus on individuals terms, the proposed method can be a useful tool to analyze sets of activated regions and predicts the paradigms performed. |
|
Paper Statistics: | Total View : 477 | Downloads : 468 | Page No: 3-8 | |
Cite this Article: | Click here to get all Styles of Citation using DOI of the article. |