Classifying schizophrenia patients and healthy individuals: Whole brain SPECT functional connectivity using support...

Area of Science:

• Neuroimaging
• Psychiatric Disorders
• Machine Learning

Background:

• Functional magnetic resonance imaging (fMRI) is widely used for brain network analysis in disorders like schizophrenia.
• Single photon emission computed tomography (SPECT) also measures neural activity via blood flow and tracers.
• Limited research exists on using SPECT data for individual subject classification in schizophrenia.

Purpose of the Study:

• To evaluate the accuracy of individual diagnostic prediction for schizophrenia using SPECT data.
• To compare the performance of various machine learning classifiers on SPECT-derived functional brain networks.

Main Methods:

• Utilized spatially constrained independent component analysis (sc-ICA) on SPECT data from 213 subjects (137 schizophrenia patients, 76 controls).
• Network priors were derived from fMRI data using the NeuroMark fMRI 1.0 template.
• Initially employed a support vector machine (SVM) classifier, followed by post hoc evaluation of random forest, logistic regression, voting, and multilayer perceptron.

Main Results:

• Linear SVM performed poorly, with more false negatives compared to other methods.
• Random forest achieved 88% sensitivity and 61% specificity.
• Logistic regression achieved 87% sensitivity and 68% specificity.

Conclusions:

• Random forest and logistic regression demonstrate superior performance for schizophrenia classification using SPECT data.
• These findings highlight the potential of sc-ICA combined with advanced classifiers for understanding aberrant functional networks in schizophrenia.
• Further investigation into these methods is recommended for future SPECT studies.