Exploring the impact of muti-site data acquisition parameters on deep learning models for fMRI data

1.  Our results demonstrate that multisite datasets, such as ABIDE-I in this case, can still show site specific effects even after data preprocessing, which can lead to suboptimal classification results.
2. Our results also show that the machine-learning models may be learning site-specific variations in the patterns that may not be specific to the ASD biomarkers and requires further investigation.