XGBoost vs. TabPFN in Neuroimaging Machine Learning-based analysis

Machine learning (ML) and deep learning (DL) applications have gained popularity in the field of neuroimaging in recent years. Here, we present a comparison between a state-of-the-art gradient boosting technique, the extreme gradient boosting (XGBoost), and a recently developed DL method, TabPFN, to assess the prediction of cognitive deficit in a large pathological population through structural and functional MRI markers. Overall, our results showed that conventional ML might still be the preferable choice for noisy tabular datasets (like neuroimaging data), also for their better explainability.