作者: Jungsun Lee , Myong-Wuk Chon , Harin Kim , Yogesh Rathi , Sylvain Bouix
DOI: 10.1016/J.NICL.2018.02.007
关键词:
摘要: Abstract Objectives Many studies have attempted to discriminate patients with schizophrenia from healthy controls by machine learning using structural or functional MRI. We included both and diffusion MRI (dMRI) performed random forest (RF) support vector (SVM) in this study. Methods evaluated the performance of classifying RF method SVM 504 features (volume and/or fractional anisotropy trace) 184 brain regions. enrolled 47 23 age- sex-matched resampled our data into a balanced dataset Synthetic Minority Oversampling Technique method. randomly permuted classification all participants as patient control 100 times ran leave one out cross validation for each permutation. then compared sensitivity specificity original dataset. Results Classification showed significantly higher rate chance: (87.6% vs. 47.0%) (95.9 48.4%) RF, (89.5% 48.0%) (94.5% 47.1%) SVM. Conclusions Machine volume measures can high degree performance. Further replications are required.