Assessing the Impact of Preprocessing Pipelines on fMRI Based Autism Spectrum Disorder Classification: ABIDE II Results

Resting-state functional MRI (rs-fMRI), a tool for assessing the brain's spontaneous activity, plays a crucial role in understanding functional connectivity, contingent on the precision of Blood Oxygen Level Dependent (BOLD) signal processing. At the forefront of this process lies preprocessing as a fundamental step in the analysis of resting-state fMRI data, enabling any subsequent investigations. This research focuses on assessing the impact of three distinct preprocessing methods on the classification of resting-state fMRI data, utilizing a range of classifiers including Support Vector Classifier with radial basis function (SVCrbf), Linear Support Vector Classifier (LinearSVC), ridge, K-Nearest Neighbors (KNN), Logistic Regression (LR), Decision Trees (DT), Random Forests (RF), and Adaptive Boosting (AdaBoost). The objective is to understand how the order and efficiency of these preprocessing steps influence the classification of Autism spectrum disorder (ASD). We conduct standard preprocessing steps on the fMRI data, including slice-timing, realignment, segmentation, co-registration, normalization, and smoothing methods. Additionally, the brain was parcellated into AAL and CC200 atlases. The evaluation, involving 1076 subjects from the ABIDE II database, represents the first application of these preprocessing methods to this dataset. Results reveal that the choice and the order of preprocessing steps significantly impact the ability to classify ASD accurately. Notably, the preprocessing strategy involving dropping the first 10 volumes, realignment, slice timing, normalization, and smoothing, yielded the best accuracy with the Ridge classifier and AAL atlas, achieving an accuracy of 65.42%, specificity of 70.73%, and AUC of 68.04%. The findings highlight the significant impact of selected preprocessing methods on the accuracy of functional connectivity classifications, underlining the importance of strategic method selection to achieve the most favorable outcomes in ASD classification.