Uncertainty Quantification in Alzheimer's Disease Progression Modeling

TL;DR

This work tackles the reliability of prognostic forecasts in Alzheimer’s disease by explicitly modeling uncertainty in 4-year MMSE trajectories. It compares four uncertainty quantification approaches—Monte Carlo Dropout, Variational Inference, Monte Carlo Markov Chain, and Ensemble Learning—within a Bayesian-inspired framework on a 512-patient ADNI-TADPOLE dataset with multimodal features. The study finds MC Dropout and MCMC to yield well-calibrated, accurate predictions under noisy data, while VI and Ensemble show calibration limitations or overconfidence, highlighting the practical trade-offs between computational efficiency and uncertainty quality. The findings support deploying MC Dropout as a robust, data-efficient method for reliable long-range cognitive trajectory forecasting in clinical settings, with implications for deploying uncertainty-aware prognostic tools in AD care.

Abstract

With the increasing number of patients diagnosed with Alzheimer's Disease, prognosis models have the potential to aid in early disease detection. However, current approaches raise dependability concerns as they do not account for uncertainty. In this work, we compare the performance of Monte Carlo Dropout, Variational Inference, Markov Chain Monte Carlo, and Ensemble Learning trained on 512 patients to predict 4-year cognitive score trajectories with confidence bounds. We show that MC Dropout and MCMC are able to produce well-calibrated, and accurate predictions under noisy training data.

Figures (15)

• Figure 1: A high level representation of the input variables and the proposed Bayesian Setup
• Figure 2: The percentage of missing value per feature in the filtered dataset used for imputation.
• Figure 3: The percentage of missing value per feature used for training.
• Figure 4: MC Dropout Calibration.
• Figure 5: VI Calibration.