MRI Image Segmentation for Schizophrenia Diagnosis Using a Modified U-Net Architecture

In order to effectively manage the progressive nature of schizophrenia, accurate characterization of the Lateral Ventricles in MRI scans is essential. Clinicians often face challenges in detecting ventricles due to potential blending with background pixels. Moreover, the clinician’s cognitive state significantly impacts accurate identification of affected brain regions. The aim of the study involves automated segmentation of lateral ventricles in the Schizophrenia using Modified U-Net method. We trained and validated the Modified U-Net using Open Neuro datasets, tuning hyperparameters for optimal performance. The adam optimizer produced better accuracy of 94.56% compared to RMS in U-Net segmentation. Our results show an Mean Intersection over Union (mIoU) of 0.9198, and mF1 of 0.9342 in U-Net segmentation. The use of the Modified U-Net to localize lateral ventricles in medical imaging has the potential to enhance schizophrenia early diagnosis.