Role of fMRI Denoising for Classification of Schizophrenia from Functional Brain Connectivity

Abstract

This study explores the impact of denoising strategies on classifying first-episode psychosis (FEP) patients from healthy controls using functional connectivity measures derived from fMRI data. Leveraging a dataset of 100 FEP patients and 90 healthy controls, the research evaluates how different preprocessing approaches—ranging from raw data to moderate and stringent denoising – affect the classification accuracy when applying logistic regression on dimension-reduced features via PCA. The findings reveal that both moderate and stringent denoising methods significantly enhance classification performance compared to using raw data, with moderate denoising reaching an 82 % accuracy with 24 principal components and stringent denoising achieving 81 % accuracy with 45 components. The study underscores the importance of denoising in improving the reliability of functional connectivity measures for schizophrenia classification. However, it also suggests that the choice between moderate and stringent denoising may not be critical, as combining multiple strategies did not substantially improve performance. This research highlights the potential of optimized fMRI data preprocessing in psychiatric diagnosis, providing insights into the neurodevelopmental and neurodegenerative processes underlying schizophrenia.