Study On The Classification And Diagnosis Of Depression Based On Radiomics

Major depressive disorder(MDD)is an emotional dysfunction,has a significant impact on quality of life and social economic burden;The incidence is high in China,with a lifetime prevalence of 3.4%.Subthreshold depression(St D),also known as subsyndromic depression or subclinical depression,is a clinically related depressive symptom,without meeting criteria for a full-blown major depressive disorder(MDD).Since the pathogenesis of MDD and St D is not clear at this stage,early diagnosis of MDD and St D is particularly important.Common clinical diagnosis of MDD and St D clinical manifestations and assessment scales,the most common scale is the Center for Epidemiologic Studies Depression Scale(CES-D)and 17-item Hamilton Depression Rating scale(HAMD-17).Due to the scale test is easy to be affected by the subjective factors of doctors and patients,with the rapid development of neuroimaging technology in recent years,computer-aided diagnosis method has become a hot topic of research.At present,computer-aided diagnostic methods are mainly based on conventional parameters(such as gray matter volume,regional homogeneity and amplitude of low-frequency oscillations),or based on machine learning or deep learning methods to build diagnostic models.However,there are some disadvantages in the above methods: for example,the features used in the model are low-level features of the original,and its diagnostic accuracy is not high;Although the method based on deep learning is more accurate,additional depth-based learning methods,although higher accuracy can be obtained,it is a black box and the research process cannot know the internal situation,resulting in the loss of a large amount of image data.In order to solve the above problems,the main method of the present study is to use a medical image feature extraction and analysis: Radiomics.The feature extracted by this method contains a lot of texture information,is now widely used in the field of oncology but seldom used in neurological and psychiatric disorders.In this study,40 patients of MDD,57 patients of St D and 74 of NCs wereincluded.Firstly,the brain structure between MDD,St D and NCs was preliminarily discussed by the Voxel-Based Morphometry,and the gray matter volume of the whole brain was extracted for classification.Then,the basic framework of radiomics method is introduced,and the important role of radiomics features in the classification and diagnosis of MDD,St D and NCs is explored.Meanwhile,Spearman correlation coefficient was used to screen out the features significantly correlated with clinical diagnosis score for the top ten features.Finally,considering the relationship of feature vectors between brain regions as a new feature,the classification model was constructed by adding them to the radiomics feature.In conclusion,in this study,radiomics features were applied to the computer-aided diagnosis of MDD and St D,and the accuracy was not only higher than that of the original classification method of low-order features,but also the location of recognized features could be used to identify disease lesions.It provides a new perspective for the follow-up study on the pathological mechanism of MDD and St D.