Long-Term Alzheimers Disease Prediction: A Novel Image Generation Method Using Temporal Parameter Estimation with Normal Inverse Gamma Distribution on Uneven Time Series

TL;DR

This work tackles long-term Alzheimer's disease prediction from MRI under irregular time intervals by introducing a temporal parameterization of the Normal Inverse Gamma distribution (tNIG). The approach combines coordinate-neighborhood based texture and deformation feature extraction (TTCN and TDCN) with a fused local/global tNIG framework and explicit uncertainty estimation to maintain disease-related characteristics over time. Empirical results on ADNI-1 and ADNI-2 show state-of-the-art performance in both short- and long-term image prediction and disease-state classification, with robustness to missing data and uneven sampling. The proposed method offers a principled and practical path toward reliable, image-based forecasting of AD trajectories in clinical settings.

Abstract

Image generation can provide physicians with an imaging diagnosis basis in the prediction of Alzheimer's Disease (AD). Recent research has shown that long-term AD predictions by image generation often face difficulties maintaining disease-related characteristics when dealing with irregular time intervals in sequential data. Considering that the time-related aspects of the distribution can reflect changes in disease-related characteristics when images are distributed unevenly, this research proposes a model to estimate the temporal parameter within the Normal Inverse Gamma Distribution (T-NIG) to assist in generating images over the long term. The T-NIG model employs brain images from two different time points to create intermediate brain images, forecast future images, and predict the disease. T-NIG is designed by identifying features using coordinate neighborhoods. It incorporates a time parameter into the normal inverse gamma distribution to understand how features change in brain imaging sequences that have varying time intervals. Additionally, T-NIG utilizes uncertainty estimation to reduce both epistemic and aleatoric uncertainties in the model, which arise from insufficient temporal data. In particular, the T-NIG model demonstrates state-of-the-art performance in both short-term and long-term prediction tasks within the dataset. Experimental results indicate that T-NIG is proficient in forecasting disease progression while maintaining disease-related characteristics, even when faced with an irregular temporal data distribution.

Figures (5)

• Figure 1: The T-NIG model employs brain images from two different time points to create intermediate brain images, forecast future images, and predict the disease. The original MRI scans $I_{0}$ and $I_{2}$ were segmented to remove the skull, followed by normalization. T-NIG extracts disease-related features from two preprocessed brain MRI images $I_{0}$ and $I_{2}$, using the TTCN and TDCN modules. The extracted information is then fed into the tNIG module for parameter fusion and parameter uncertainty estimation, thereby enabling the prediction and generation of brain images $I_{1}$,$I_{t-1}$, $I_{t}$.
• Figure 2: Long-term brain image generation in ADNI-1. Each model utilizes brain images of subjects as input at ages 71 and 77.The brain images for ages 74 and 76 are generated through interpolation, whereas the brain images for ages 78 to 87 are predictions. In each group of images, the first row displays the predicted brain images, while the second row shows the differences between the predicted and actual brain images.
• Figure 3: A detailed comparison of the three planes of the interpolated brain images and the predicted brain images in ADNI-1 generated by various methods is presented. The input images are from a 71-year-old and a 77-year-old individual, and output images are 74 and 84. Axial, Coronal and Sagittal represent the three planes in brain image.
• Figure 4: Brain image generation across different age groups. The first row represent the three planes in brain image; The second row illustrate the ages of the input and output brain images; The first row of each group displays the brain images generated by the corresponding models, while the second row of each group of images represents the differences between the generated brain images and the original images.
• Figure 5: The experiment evaluates how well the T-NIG distribution, T distribution, Exponential distribution, and Laplace distribution fit the changes in features among brain images. The sequence of brain imaging follows the pattern illustrated in Figure \ref{['Figure_3']}.