基于影像组学的重度抑郁症及阈下抑郁症分类研究

Classification Study of Major Depressive Disorder and Subthreshold Depression Based on Radiomics

目的识别与抑郁症诊断相关的影像组学特征,并基于所识别的特征建立和评估重度抑郁症(MDD)和阈下抑郁症(StD)的分类模型。资料与方法纳入171例受检者,其中MDD 40例、StD 57例、正常对照组74例,进行头部MRI扫描,获得T1WI图像,基于解剖图谱提取116个脑区的影像组学特征,通过基于树的特征选择方法计算初始特征的重要程度,在已识别特征的基础上构建SVM模型,并通过准确度、敏感度和特异度评估分类器性能。结果使用分类器区分MDD患者与正常对照组的分类准确度为86.51%,使用分类器区分StD患者和正常对照组的准确度为72.74%。对照组与MDD组分析显示,重要程度排名前10位的特征中所占比例最高的脑区位于颞极;而StD组与正常对照组分析显示,所占比例最高的脑区位于小脑。结论基于影像组学的方法在诊断分类MDD和StD方面具有潜在的效用,通过对识别特征的位置进行定位,可以对疾病的病灶进行识别。

Purpose To identify radiomics features related to diagnosis of depression,then to build and evaluate classification models for major depressive disorder(MDD)and subthreshold depression(StD)based on identified features.Materials and Methods 171 subjects were enrolled in this study,including 40 patients with MDD,57 patients with StD and 74 patients with normal controls.MRI scans of the brain were performed to obtain T1WI images.The radiomic features of 116 brain regions was extracted based on the automatic anatomical labeling.The tree-based feature selection method was used to calculate the importance degree of the initial feature.Support vector machine classifier was constructed on the basis of the identified features,and the performance of the classifier was evaluated by the accuracy,sensitivity and specificity.Results The classifier was used to distinguish the classification accuracy of MDD patients from the normal control group by 86.51%.The accuracy of using the classifier to distinguish StD patients from the normal control group was 72.74%.When the control group and the MDD group were analyzed,the regions with the highest proportion of the 10 most important features were located in the temporal pole,while the regions with the highest proportion of the StD group and the normal control group were located in the cerebellum.Conclusion Radiomics has a potential utility in classifying MDD and StD.The lesions of the disease can be identified by locating the location of the identified features.