Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TE-SSL: Time and Event-aware Self Supervised Learning for Alzheimer's Disease Progression Analysis

Jacob Thrasher, Alina Devkota, Ahmed Tafti, Binod Bhattarai, Prashnna Gyawali

TL;DR

The paper addresses Alzheimer’s disease progression analysis by introducing TE-SSL, a time- and event-aware self-supervised learning framework that leverages both event indicators and time-to-event labels to guide representation learning. TE-SSL extends contrastive SSL with a time-aware weighting scheme $\omega_{i,j}$ based on the time differences $\Delta_{i,j}$, emphasizing pairs at similar progression stages, and couples this with a DeepHit-based downstream head to predict a distribution of hazards over discrete times via $\mathcal{L}_{total} = \mathcal{L}_1 + \mathcal{L}_2$. Empirical results on the ADNI dataset show TE-SSL outperforms baselines in time-to-event metrics such as $C\text{-}td$ and IBS, with ablations confirming the benefit of jointly using event and time-to-event supervision. The approach demonstrates practical utility in learning progression-aware MRI representations that enhance prognostic performance in Alzheimer's disease.

Abstract

Alzheimer's Dementia (AD) represents one of the most pressing challenges in the field of neurodegenerative disorders, with its progression analysis being crucial for understanding disease dynamics and developing targeted interventions. Recent advancements in deep learning and various representation learning strategies, including self-supervised learning (SSL), have shown significant promise in enhancing medical image analysis, providing innovative ways to extract meaningful patterns from complex data. Notably, the computer vision literature has demonstrated that incorporating supervisory signals into SSL can further augment model performance by guiding the learning process with additional relevant information. However, the application of such supervisory signals in the context of disease progression analysis remains largely unexplored. This gap is particularly pronounced given the inherent challenges of incorporating both event and time-to-event information into the learning paradigm. Addressing this, we propose a novel framework, Time and Even-aware SSL (TE-SSL), which integrates time-to-event and event data as supervisory signals to refine the learning process. Our comparative analysis with existing SSL-based methods in the downstream task of survival analysis shows superior performance across standard metrics.

TE-SSL: Time and Event-aware Self Supervised Learning for Alzheimer's Disease Progression Analysis

TL;DR

The paper addresses Alzheimer’s disease progression analysis by introducing TE-SSL, a time- and event-aware self-supervised learning framework that leverages both event indicators and time-to-event labels to guide representation learning. TE-SSL extends contrastive SSL with a time-aware weighting scheme based on the time differences , emphasizing pairs at similar progression stages, and couples this with a DeepHit-based downstream head to predict a distribution of hazards over discrete times via . Empirical results on the ADNI dataset show TE-SSL outperforms baselines in time-to-event metrics such as and IBS, with ablations confirming the benefit of jointly using event and time-to-event supervision. The approach demonstrates practical utility in learning progression-aware MRI representations that enhance prognostic performance in Alzheimer's disease.

Abstract

Alzheimer's Dementia (AD) represents one of the most pressing challenges in the field of neurodegenerative disorders, with its progression analysis being crucial for understanding disease dynamics and developing targeted interventions. Recent advancements in deep learning and various representation learning strategies, including self-supervised learning (SSL), have shown significant promise in enhancing medical image analysis, providing innovative ways to extract meaningful patterns from complex data. Notably, the computer vision literature has demonstrated that incorporating supervisory signals into SSL can further augment model performance by guiding the learning process with additional relevant information. However, the application of such supervisory signals in the context of disease progression analysis remains largely unexplored. This gap is particularly pronounced given the inherent challenges of incorporating both event and time-to-event information into the learning paradigm. Addressing this, we propose a novel framework, Time and Even-aware SSL (TE-SSL), which integrates time-to-event and event data as supervisory signals to refine the learning process. Our comparative analysis with existing SSL-based methods in the downstream task of survival analysis shows superior performance across standard metrics.
Paper Structure (15 sections, 6 equations, 2 figures, 2 tables)

This paper contains 15 sections, 6 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. TE-SSL: Time and Event-aware Self Supervised Learning
  4. Self-supervised learning
  5. Supervisory signal for SSL training
  6. Time-to-event and Event labels for SSL training
  7. Time-to-event prediction
  8. Experiments and Results
  9. ADNI dataset:
  10. Implementation details:
  11. Results
  12. Ablation analysis
  13. Conclusion
  14. Acknowledgments.
  15. Disclosure of interests.

Figures (2)

  • Figure 1: Schematic diagram of the proposed time- and event-aware SSL, where $\Delta_{*,*}$ represents the time difference between two data elements $z_*$
  • Figure 2: t-SNE analysis of representations across different SSL frameworks. Individual points, if not censored, are labeled with different time-to-event groups.