Exploring Reproducibility of fMRI Analysis Using SPM and Nilearn Open-Source Tools

Functional magnetic resonance imaging (fMRI) is a widely used technique for studying human brain activity. Various libraries and tools are available for performing the analysis and preprocessing of data. However, a significant challenge is that these different tools often yield varying results, which can complicate the interpretation and comparison of studies. In this context to contribute to the transparency of the used framework and reproducibility of the results in fMRI studies, this work presents a comparative analysis using Nilearn, a library available for Python, and SPM, a graphical interface available in MATLAB, along with preprocessing conducted with SPM in Nipype, also available for Python. Blood oxygen level- dependent (BOLD) contrast and anatomical volumes from one subject of the Flanker Task dataset were used. For this purpose, three distinct contrasts were established: Congruent > Incongruent, Congruent > Baseline, and Incongruent > Baseline. A point-by-point comparison of these masks was performed using a density plot and visualization of the different masks in the subject's anatomical volume. The results indicated that the design matrices of SPM and Nilearn are slightly different despite being configured with the same parameters, leading to significant differences in the final results.