Machine Learning Approaches for the Detection of Schizophrenia Using Structural MRI

The reproducibility of Computer Aided Diagnosis (CAD) in detecting schizophrenia using neuroimaging modalities can provide early diagnosis of the disease. Schizophrenia is a psychiatric disorder that can lead to structural abnormalities in the brain, causing delusions and hallucinations. Neuroimaging modality such as a structural Magnetic Resonance Imaging (sMRI) technique can capture these structural abnormalities in the brain. Utilizing Machine Learning (ML) as a potential diagnostic tool in detecting classification biomarkers can aid clinical measures and cater to recognizing the factors underlying schizophrenia. This paper proposes an ML based model for the detection of schizophrenia on the structural MRI dataset of 146 subjects. We sought to classify schizophrenia and healthy control using five ML classifiers: Support Vector Machine, Logistic Regression, Decision Tree, k-Nearest Neighbor, and Random Forest. The raw structural MRI scans have been pre-processed using techniques such as image selection, image conversion, gray scaling of MRI images, and image flattening. Further, we have tested the performance of the model using hold-out cross-validation and stratified 10-fold cross-validation techniques. The results showed that the SVM achieved high accuracy when the dataset was validated using a stratified 10-fold cross-validation technique. On the other hand, k-Nearest Neighbor performed better when the hold-out validation method was used to evaluate the classifier.