Accuracy of Deep Learning Algorithms in the Classification of Schizophrenia Patients Using MRI Images


1 Tabriz University of Medical Sciences <>

2 Radiology Department ParaMedical Faculty, Tabriz University of Medical Sciences <>



Schizophrenia is one of the acute mental disordersthatdisturbpatients’cognition,behavior and emotion. Its clinical diagnosis is usually based on fulfilling criteria of phenotypical features, which is time-consuming. Therefore, using the methods for early diagnosis of schizophrenia and faster therapeutic interventions is essential. Machine learning algorithms have shown good performance for schizophrenia classification, but due to the limitation in manual feature selection, they may not fully represent the neural differences associated with schizophrenia. On the other hand, deep learning algorithms, especially convolutional neural networks (CNN), can learn the fully automatic features related to schizophrenia. The aim of this study is to determine the accuracy of the deep learning algorithms in the classification of schizophrenia patients using MRI images.
We searched the articles published in 2018-2022 in PubMed, Google Scholar, and AltaVista databases using the keywords of schizophrenia, classification, deep learning algorithms, and magnetic resonance imaging. Among the searched articles, the most relevant ones were reviewed.
According to the findings, various deep learning algorithms such as pre-trained 2D CNN, 3D Naïve CNN, modified 3D VGG with squeeze excitation (SE) and batch normalization (BN) model (SE-VGG-11BN), and 2D convolutional Autoencoder (CNN-AE) have been used for the classification of schizophrenia using structural, diffusion, and functional MRI images. The reviewed articles results showed an accuracy
of 72-97% for classifying schizophrenia in MRI images using these algorithms.
Deep learning algorithms showed high accuracy for the classification of schizophrenia patients using MRI images.
Keywords: Schizophrenia, Classification, Deep learning algorithms, Magnetic resonance imaging