Research On Attention Deficit Hyperactivity Disorder Classification Based On Machine Learning

Attention deficit hyperactivity disorder(ADHD)is a common mental disorder in childhood.At present,the diagnosis of ADHD is mostly in the form of questionnaires and scales,with a large degree of subjectivity.The objective diagnosis method of ADHD is mainly based on magnetic resonance imaging(MRI)data.Clinicians diagnose ADHD by screening MR images repeatedly.However,because the difference of MR images between normal people and ADHD patients is very insignificant,it is difficult to distinguish accurately,and misdiagnosis and missed diagnosis are easy to occur.Therefore,in this paper,the MR image and machine learning method are used to realize the automatic discrimination of ADHD images,and an ADHD image classification method based on machine learning is proposed.The main research work is as follows:(1).ADHD classification based on classical machine learning.Support vector machine and neural network algorithm for ADHD classification are researched in this paper.The experimental data set comes from the ADHD-200 contest database and cross-validation is used to train and test the model.(2).ADHD classification based on ROI-CNN model.Due to the high similarity of image intensities in different brain patterns,an effective region of interest(ROI)extraction strategy is introduced for initial feature selection.This paper mainly researched the ROI fine segmentation method based on level set and region growth(LS-RG)and the ROI rough segmentation method based on improved template matching,and achieve a good segmentation effect on the ADHD-200 competition data set.Convolutional neural network(CNN)has a powerful learning ability to analyze the cognitive function of brain more deeply.In this paper,a method based on 2D-CNN is proposed to achieve ADHD feature extraction and classification.The model is verified and optimized by constructing different CNN models and using data amplification to prevent over-fitting.(3).ADHD classification based on 3D-CNN model.In order to effectively extract the depth spatial information of ADHD,this paper constructed a 3D-CNN model and successfully applied it to the automatic diagnosis of ADHD.The above experimental results show that the methods based on machine learning can realize the automatic classification of MR images between ADHD patients and normal people.Among them,the ROI-CNN method has the highest classification accuracy and the strongest model generalization ability.