Assessment of severity in neuropsychiatric disorders based on radiomic features with prior shape level set and metaheuristic algorithms

Abstract

AbstractThe diagnostic and clinical overlap between schizophrenia (SZ) and schizoaffective (SA) is vital oncological issue in psychiatry disorder. This study is designed to examine tissue variation in cerebellum region. Shape prior levelset method is used to segment cerebellum from MR image, and is subdivided by symmetry line detection. Radiomic features are extracted and subjected to Binary PSO (BPSO) fuzzy‐SVM classifier. This work is evaluated on COBRE database. The results show that, shape prior levelset could segment cerebellum region with better similarity values. The Laplacian of Gaussian and structure tensor features extracted from right cerebellum gives high significance between SA and SZ (P < 0.00001). The BPSO‐FSVM could classify the normal, SA, and SZ with an accuracy of 88.33%. The AUC is 0.913 for the combined features extracted from right cerebellum. It is concluded that optimized feature selection‐based classifier gives better results. This study is clinically helpful in the investigation of psychiatry disorder.