Comparation of fMRI feature extraction and classification method

Abstract

Classification of brain images obtained through functional magnetic resonance imaging (fMRI) poses a serious challenge to pattern recognition and machine learning due to the extremely large feature-to-instance ratio. This calls for revision and adaptation of the current state-of-the-art classification methods. Thus an effective application of machine learning techniques to automatically analyze such data is called for. We investigate the suitability of Support Vector Machine and Convolutional Neural Network method for fMRI classification and find that they are all effective. In this paper, we proposes a hybrid model that integrates the synergy of two superior classifiers: convolutional neural networks (CNNs) and support vector machines (SVMs), which have proven results for recognizing different types of patterns. In this model, the CNN works as a trainable feature extractor and the SVM functions as a recognizer. This hybrid model automatically extracts features from raw images and generates predictions. We conducted experiments on Haxby's 2001 fMRI dataset. Comparisons with Haxby’s study using the same database indicate that the proposed fusion achieved better results: a classification accuracy of 99.5%. And we compared this model with other popular algorithm to evaluate our model's effectivity.