TabPFN Outperforms Traditional ML for MCI-to-AD Conversion Prediction

A new study evaluates TabPFN, a tabular pre-trained foundation network, for predicting conversion from Mild Cognitive Impairment (MCI) to Alzheimer's Disease (AD) within three years. Using the TADPOLE dataset from ADNI, researchers compared TabPFN against XGBoost, Random Forest, LightGBM, and Logistic Regression across training set sizes from 50 to 1000 samples. Multimodal biomarkers included demographics, APOE4, MRI volumes, CSF markers, and PET imaging. TabPFN achieved the highest AUC of 0.892, outperforming LightGBM (0.860), and showed particular strength in low-data settings, maintaining strong AUC with only 50 training samples. The findings suggest TabPFN is a promising tool for early AD intervention in data-limited clinical environments.