Research On Feature Selection And Classification Of Brain Image Based On Elastic-Net

Autistic Spectrum Disorders(ASD)is a syndrome with social fear and stereotyped behavior.There is still no reliable biomarker,and it is a syndrome defined by behavior.Although many studies have focused on detecting clinically useful biomarkers based on Resting-state functional magnetic resonance imaging,they have not yet established a solid population in the feature selection process relationship,resulting in a lot of information loss.Based on the large multi-center Autism Brain Imaging Data Exchange(ABIDE)database,the resting-state functional magnetic resonance imaging is studied,combined with the Lasso method and Adaptive Lasso method for feature selection,but the experimental results show that the model predicts.The effect needs to be improved.To better identify diseases with small sample size,high data dimensions,and many influencing factors,and to better assist clinical diagnosis decision-making,we propose a method based on resting-state functional magnetic resonance imaging combined with Elastic-Net method for feature selection,and finally,realize the predictive diagnosis of autism spectrum disorder.The key research and innovations of this thesis are as follows:(1)In the feature selection process,when the data is not processed in any way,redundant information cannot be eliminated and most of the original information is retained,which makes the obvious features between normal subjects and ASD-affected subjects blurred.It is proposed to threshold the data of the subjects first to retain some of the positively correlated features with higher synchronization,and then use the Lasso regression method for feature selection.(2)In the process of feature selection using the Lasso method,it still imposes the same penalty on features of different importance.It is proposed to apply the improved adaptive Lasso method to the model.The experimental results show that its performance predicts the index value.Compared with the Lasso method,it is improved.(3)Regarding the Lasso method and its related improvement methods,when solving the problem of grouping effects,they tend to choose a feature and ignore other features,which may lose a lot of key information,and the model effect is not satisfactory.The resting-state functional magnetic resonance imaging-based on the human brain is proposed,combined with the Elastic-Net method to be applied to the early diagnosis and prediction of autism,and Control the experimental effect with Lasso method and Adaptive Lasso method.As an effective regression method,the Elastic-Net method proposed in this thesis can express the features in a sparse form in the original feature set.The advantage is that there is no need to perform feature selection in advance,which greatly saves running time and improves the efficiency of the algorithm.It can effectively avoid the shortcomings of the related methods mentioned in Chapter 3 and has a better forecasting effect.Finally,with the help of Support Vector Machine(SVM)to classify ASD subjects.The classification performance of the algorithm is mainly evaluated according to the classification accuracy rate,sensitivity,specificity,positive detection rate,negative detection rate and other indicators.The experimental results show the effectiveness and practicality of predictive diagnosis of ASD based on Elastic-Net method.