Classification of schizophrenia using feature-based morphometry.

Area of Science:

• Neuroimaging
• Machine Learning in Medicine
• Psychiatric Disorders

Background:

• Schizophrenia diagnosis relies on clinical assessment, lacking objective biomarkers.
• The dorsolateral prefrontal cortex (DLPFC) shows structural alterations in schizophrenia.
• Automated classification methods can aid in objective diagnosis.

Purpose of the Study:

• To develop and evaluate an automated method for classifying schizophrenia patients.
• To utilize Scale Invariance Feature Transform (SIFT) and Support Vector Machine (SVM) for classification.
• To investigate the efficacy of using the dorsolateral prefrontal cortex (DLPFC) as a region of interest (ROI).

Main Methods:

• Landmark detection and SIFT description of the DLPFC from 1.5T MRI scans.
• Feature vocabulary construction and Bag-of-Words (BoW) model for brain representation.
• Non-linear SVM classification with a local kernel and a novel weighting approach for feature relevance.

Main Results:

• The combined ROI-SVM approach achieved classification accuracies of 75% (left DLPFC) and 66.38% (right DLPFC).
• Higher classification performance was observed for females (up to 84.09%) and seniors (up to 81.25%) when analyzed separately.
• Supervised weighting functions generally improved classification efficacy across analyses.

Conclusions:

• The integrated ROI-SVM approach reliably detects schizophrenia based on DLPFC structural markers.
• This method offers a promising tool for objective schizophrenia detection.
• Future studies should focus on first-episode patients, analyzing males and females separately.