Image normalization can improve the effectiveness of radiomics model based on multiscale structural MRI for diagnosing MDD and StD in adolescents

We developed a radiomics classifier for MDD and StD in adolescents with multiscale structural MRI after normalization, and it had the best performance and was superior to the classifier based on conventional image indicators and unnormalized radiomics features. The AUC, sensitivity, and accuracy for discriminating MDD and HC, MDD and StD, StD and HC were 0.928, 89.2% and 90.5%, 0.821, 73.0% and 80.8%, 0.836%, 82.4% and 79.7% respectively. The high discriminant radiomics features of cuneiform lobe and cerebellum (lobule ⅵ, ⅶ-b and ⅹ, 4/5 area of cerebellar vermis) played a key role in the pathophysiological mechanism of MDD and StD.