Research Of Feature Selection Method Base On MRI Data

Feature selection is an important topic of pattern recognition for enhancing classification and potential biomarker discovery in medical image analysis. However, traditional multivariate methods is likely to obtain unstable and unreliable results in case of an extremely high dimensional feature space and very limited training samples, where the features are often correlated or redundant.In order to improve the interpretations of the discovered potential biomarker and the robustness of the resultant classifier, this paper will focus on the feature selection method of f MRI data: after analyzing the characteristics of MRI data, we introduced a novel feature selection method which combines a recent implementation of the stability selection approach and the elastic net approach, and we proved this method has the great robustness than the traditional multivariate method through a simulation test; and we find this method has the advantage in the data analysising of mutil-center network, through a data processing of multi-center attentiondeficit/hyperactivity disorder(ADHD) f MRI data; and through a data analysising of a face data, we find the method is effective to the structure f MRI data. Some aspects of this dissertation have been put forward:1 We introduced a novel feature selection method which combines a recent implementation of the stability selection approach and the elastic net approach, after analyzing the characteristics of MRI data. And through a simulation experiment, we find SS-EN can get a better result than traditional mutilvariate method in the strong noise interference, which means the SS-EN has a great robustness.2 We used a multi-center ADHD data to build a prediction model, find the potential biomarker,then used to classify. The result proved our method is useful for real f MRI data, not only the biomarker have the outstanding physiological significance, but the classification performance is better than the other method.3 We used a face recognition data to find the core regions of face recognition. The result proved our method is useful for detecting of the activity region of the cognitive task, and compared with other method, SS-EN can find the most complete and spatially continuous regions.