Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Individualized multi-horizon MRI trajectory prediction for Alzheimer's Disease

Rosemary He, Gabriella Ang, Daniel Tward

TL;DR

Alzheimer's disease MRI biomarkers are limited by high intersubject variability. The paper introduces a double-encoder CVAE conditioned on a base image, age, disease status, and time difference to generate multi-horizon, high-resolution 3D MRI predictions for individual patients, with a latent space of dimension $10$ for sampling. Evaluations on ADNI and external OASIS data show improved trajectory accuracy, the ability to interpolate and extrapolate up to 10 years, and a latent-space posterior classifier for disease status. The approach enables early diagnosis support, treatment effect estimation, and the creation of synthetic data for downstream tasks, while noting limitations such as the lack of head-to-head method comparisons and the focus on imaging and demographics.

Abstract

Neurodegeneration as measured through magnetic resonance imaging (MRI) is recognized as a potential biomarker for diagnosing Alzheimer's disease (AD), but is generally considered less specific than amyloid or tau based biomarkers. Due to a large amount of variability in brain anatomy between different individuals, we hypothesize that leveraging MRI time series can help improve specificity, by treating each patient as their own baseline. Here we turn to conditional variational autoencoders to generate individualized MRI predictions given the subject's age, disease status and one previous scan. Using serial imaging data from the Alzheimer's Disease Neuroimaging Initiative, we train a novel architecture to build a latent space distribution which can be sampled from to generate future predictions of changing anatomy. This enables us to extrapolate beyond the dataset and predict MRIs up to 10 years. We evaluated the model on a held-out set from ADNI and an independent dataset (from Open Access Series of Imaging Studies). By comparing to several alternatives, we show that our model produces more individualized images with higher resolution. Further, if an individual already has a follow-up MRI, we demonstrate a usage of our model to compute a likelihood ratio classifier for disease status. In practice, the model may be able to assist in early diagnosis of AD and provide a counterfactual baseline trajectory for treatment effect estimation. Furthermore, it generates a synthetic dataset that can potentially be used for downstream tasks such as anomaly detection and classification.

Individualized multi-horizon MRI trajectory prediction for Alzheimer's Disease

TL;DR

Alzheimer's disease MRI biomarkers are limited by high intersubject variability. The paper introduces a double-encoder CVAE conditioned on a base image, age, disease status, and time difference to generate multi-horizon, high-resolution 3D MRI predictions for individual patients, with a latent space of dimension for sampling. Evaluations on ADNI and external OASIS data show improved trajectory accuracy, the ability to interpolate and extrapolate up to 10 years, and a latent-space posterior classifier for disease status. The approach enables early diagnosis support, treatment effect estimation, and the creation of synthetic data for downstream tasks, while noting limitations such as the lack of head-to-head method comparisons and the focus on imaging and demographics.

Abstract

Neurodegeneration as measured through magnetic resonance imaging (MRI) is recognized as a potential biomarker for diagnosing Alzheimer's disease (AD), but is generally considered less specific than amyloid or tau based biomarkers. Due to a large amount of variability in brain anatomy between different individuals, we hypothesize that leveraging MRI time series can help improve specificity, by treating each patient as their own baseline. Here we turn to conditional variational autoencoders to generate individualized MRI predictions given the subject's age, disease status and one previous scan. Using serial imaging data from the Alzheimer's Disease Neuroimaging Initiative, we train a novel architecture to build a latent space distribution which can be sampled from to generate future predictions of changing anatomy. This enables us to extrapolate beyond the dataset and predict MRIs up to 10 years. We evaluated the model on a held-out set from ADNI and an independent dataset (from Open Access Series of Imaging Studies). By comparing to several alternatives, we show that our model produces more individualized images with higher resolution. Further, if an individual already has a follow-up MRI, we demonstrate a usage of our model to compute a likelihood ratio classifier for disease status. In practice, the model may be able to assist in early diagnosis of AD and provide a counterfactual baseline trajectory for treatment effect estimation. Furthermore, it generates a synthetic dataset that can potentially be used for downstream tasks such as anomaly detection and classification.
Paper Structure (18 sections, 2 equations, 5 figures, 1 table)

This paper contains 18 sections, 2 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. Background
  3. Related work
  4. Our Contribution
  5. Method
  6. Data Preparation
  7. Autoencoders, VAE and CVAE
  8. Our Double Encoder CVAE
  9. Objective function
  10. Model architecture
  11. Training procedure
  12. Comparison to alternative methods
  13. Low rank linear prediction
  14. VAE with linear mixed effect estimation
  15. Results
  16. ...and 3 more sections

Figures (5)

  • Figure 1: Model architecture.
  • Figure 2: Sample prediction in the held-out test set.
  • Figure 3: MSE (log scale) in test (left) and external (right) validation set in 3 ROIs.
  • Figure 4: Ten year prediction from one base image. Incremental changes (non-cumulative) shown via optical flow divergence, blue: expansion, red: contraction.
  • Figure 5: Hypothesis testing for disease status, $H_0$: null, $H_1$: AD.