Influence of Functional Magnetic Resonance Imaging Data Preprocessing Pipelines on the Accuracy of Schizophrenia...

Area of Science:

• Neuroimaging
• Computational Psychiatry
• Machine Learning

Background:

• Accurate classification of schizophrenia patients from healthy controls is crucial for diagnosis and treatment.
• Functional magnetic resonance imaging (fMRI) offers insights into brain activity but requires rigorous preprocessing.
• Machine learning (ML) methods are increasingly used to analyze fMRI data for clinical applications.

Purpose of the Study:

• To analyze the influence of various preprocessing pipelines on the accuracy of classifying schizophrenia patients and healthy controls using fMRI data.
• To provide recommendations for optimizing fMRI data preprocessing pipelines for schizophrenia classification tasks.

Main Methods:

• Utilized fMRI data from 72 subjects.
• Applied seven distinct preprocessing pipelines to each dataset.
• Constructed feature vectors using ReHo, FCM, and FHR algorithms.
• Performed classification using 15 ML methods from scikit-learn.

Main Results:

• No single preprocessing pipeline universally optimized all feature vector construction algorithms.
• Spatial smoothing improved accuracy for ReHo and FHR metrics but not FCM.
• Frequency filtering and median normalization substantially increased accuracy for ReHo and FCM.
• Inhomogeneity and slice timing correction improved accuracy for ReHo but not FCM.

Conclusions:

• Recommended preprocessing pipelines tailored to specific feature extraction methods (ReHo, FCM, FHR).
• Highlighted the importance of spatial smoothing for FHR and filtering/normalization for ReHo and FCM.
• Advised caution with ICA filtering due to inconsistent effects on classification accuracy.