Ventricle Analysis in MR images for Neurodegenerative and Neuropsychiatric Disorder using Optimized Radiomic and Fused classification

Dementia and schizophrenia are complex neurodisorders which require desired biomarkers to precisely differentiate between stages of severity. This work is an attempt to understand the brain left and right ventricle changes in MR images for normal, mild cognitive impairment, early MCI, MCI, late MCI, Alzheimer disease, Schizophrenia and Schizoaffective using significant radiomic features and fused classifiers. The images are obtained from public database. Then the sub-anatomic left and right ventricle regions are attained using Tsallis entropy-based thresholding and palindrome axis detection method. These segmented regions morphometric characterization is analysed with help of radiomic features. Further, significant features are selected from these regions. They are subject to SVM, KNN, NB, RF, LDA and fused classifier to differentiate the considered classes. Result suggests that left and right ventricle is prominently delineated with the average segmented accuracy of 96.4% and 97.2% respectively. The statistical test reveals that the prominent features from these regions show desired difference with p < 0.0001. Finally, the fused classifier is able to achieve the higher accuracy of 82.86% for left ventricle and 75.71% for right ventricle for considered classes. This concludes that left ventricle seems to relatively capable to distinguish the considered stages.