Evaluating TabPFN for Mild Cognitive Impairment to Alzheimer's Disease Conversion in Data Limited Settings

2026-04-29

AI Summary

This study addresses the challenge of predicting conversion from mild cognitive impairment (MCI) to Alzheimer’s disease (AD), where limited longitudinal data often undermines model reliability. For the first time, the authors introduce the Tabular Pre-Trained Foundation Network (TabPFN) to this task, integrating multimodal biomarkers—including demographic variables, APOE4 genotype, MRI volumetric measures, cerebrospinal fluid (CSF) analytes, and PET imaging features—and systematically benchmark its performance against established methods such as XGBoost, Random Forest, LightGBM, and logistic regression across varying training sample sizes. Results demonstrate that TabPFN achieves an AUC of 0.892 on the full dataset, outperforming the best conventional approach (LightGBM, AUC=0.860), and maintains strong predictive performance even with only 50 training samples, thereby highlighting the potential of foundation models in medical settings characterized by scarce data.

Abstract

Accurate prediction of conversion from Mild Cognitive Impairment (MCI) to Alzheimers Diseases (AD) is essential for early intervention, however, developing reliable conversion predictive models is difficult to develop due to limited longitudinal data availability We evaluate TabPFN (Tabular Pre-Trained Foundation Network) against traditional machine learning methods for predicting 3 year MCI to AD conversion using the TADPOLE dataset derived from ADNI. Using multimodal biomarker features extracted from demographics, APOE4, MRI volumes, CSF markers, and PET imaging, we conducted an experimental comparison across varying training set sizes (N=50 to 1000) and models including XGBoost, Random Forest, LightGBM, and Logistic Regression. TabPFN achieved one the highest performance (AUC=0.892), outperforming LightGBM (AUC=0.860) and demonstrating advantages in low data settings. At N=50 training samples, TabPFN maintained strong AUC while the traditional machine learning models struggles at small training samples. These findings demonstrate that foundation models are promising for disease prediction in data limited scenarios, such as Alzheimers diseases.

Problem

• Mild Cognitive Impairment
• Alzheimer's Disease
• Conversion Prediction
• Data Limited Settings
• Tabular Data

Innovation

• TabPFN
• foundation model
• Alzheimer's disease prediction
• data-limited setting
• MCI-to-AD conversion