A hybrid deep learning approach with ConvNeXt and vision transformer for detecting schizophrenia in fMRI scans

Abstract

Schizophrenia being one of the most complex mental disorders disrupts the default livelihood of the affected along with their close ones. Thus posing significant challenges to the individuals and the healthcare system worldwide as they often fail to diagnose the illness on time for effective treatment and management. In this case, MRI scans offer a promising opportunity for diagnosis but interpreting these complex images requires advanced analytical techniques which all physicians might not have. This study proposes a novel hybrid deep learning model composed of ConvNeXt and Vision Transformer (ViT) architectures to classify schizophrenia from 3D functional MRI. ConvNeXt is used for local feature extraction by capturing detailed and hierarchical features from the MRI images. Meanwhile, Vision Transformers make the model better at seeing the bigger picture and understanding the complex connections in the data by using a special self-attention process. Utilizing the subtle neurobiological markers in fMRI, the hybrid model will facilitate fast and proper detection of schizophrenia patients. This model demonstrates immense potential for automated schizophrenia detection by combining advanced deep learning techniques to develop precise and reliable diagnostic tools for mental healthcare.