Enhancing Brain Age Prediction and Neurodegeneration Detection with Contrastive Learning on Regional Biomechanical Properties

Jakob Träuble, Lucy V. Hiscox, Curtis L. Johnson, Angelica I. Avilés-Rivero, Carola‐Bibiane Schönlieb, Gabriele S. Kaminski SchierlebioRxiv 2025 · DOI

TL;DR

This paper introduces a novel self-supervised contrastive learning framework that leverages Magnetic Resonance Elastography (MRE) to predict brain age and detect neurodegenerative diseases. The study demonstrates that MRE, by measuring brain tissue stiffness and damping ratio, is more sensitive to age-related changes and disease progression than traditional structural MRI, offering a promising new avenue for early dementia risk detection.

What Problem Does This Paper Address?

Aging is a complex process that affects brain structure and function, leading to declines in various metrics. While traditional structural MRI has been instrumental in understanding these changes, it may not capture the subtlest, early signs of aging and neurodegeneration. Biomechanical properties of brain tissue, such as stiffness and damping ratio, which can be measured using Magnetic Resonance Elastography (MRE), are thought to be more sensitive to these early changes. However, effectively extracting and utilizing this information, especially in the context of predicting brain age and identifying subtle signs of disease, remains a challenge. This paper aims to bridge this gap by developing a sophisticated machine learning approach to harness the power of MRE for enhanced brain age prediction and early detection of neurodegenerative conditions like Mild Cognitive Impairment (MCI) and Alzheimer's Disease (AD).

Methods

The core of this research lies in a novel self-supervised contrastive regression framework. This framework is designed to predict brain age (a measure of how old a brain appears based on its imaging characteristics) from MRE-derived biomechanical properties (stiffness and damping ratio), as well as standard MRI data.

Key methodological aspects include:

  • Contrastive Regression: Unlike traditional contrastive learning for classification, this approach is adapted for regression tasks. It aims to pull samples with similar ages closer together in an embedding space and push samples with different ages further apart.
  • Adaptive Neighbourhoods: To handle non-uniform data distributions and low-data regimes, the framework dynamically adjusts the "neighbourhood" of samples considered during training. This allows the model to focus on the most relevant age-related differences.
  • Voxel-wise Analysis: The study moves beyond regional averages to analyze biomechanical properties at the voxel level, capturing finer-grained, localized aging patterns.
  • Integration of Modalities: The framework can integrate MRE-derived stiffness and damping ratio, along with standard T1-weighted MRI data, to leverage complementary information.
  • Occlusion-based Saliency Maps: To understand which brain regions are most influential for age prediction, the researchers used saliency maps, which highlight the importance of specific voxels or regions.

Datasets

The study utilized a pooled dataset comprising:

  • 311 healthy volunteers: Collected from five independent studies with similar MRE acquisition protocols.
  • 31 patients: Including 20 individuals with Mild Cognitive Impairment (MCI) and 11 individuals diagnosed with Alzheimer's Disease (AD).

All studies used a 3T MRI scanner and similar MRE acquisition parameters. This dataset is noted as one of the largest collections of MRE-derived brain mechanical properties to date.

Key Insights and Ideas

  • MRE's Superior Sensitivity: Biomechanical properties (stiffness and damping ratio) derived from MRE are significantly more sensitive to age-related brain changes than traditional structural MRI. This allows for earlier detection of subtle alterations.
  • Stiffness vs. Damping Ratio: Stiffness appears to be a dominant biomarker for later-stage neurodegeneration (AD), while damping ratio is more effective at detecting early changes associated with MCI.
  • Deep Brain Structures as Key Biomarkers: The caudate and thalamus are identified as crucial age-sensitive structures, playing a significant role in brain aging. The thalamus is particularly implicated in AD, while the hippocampus is highlighted in MCI.
  • Identifying "At-Risk" Healthy Individuals: The framework can identify cognitively healthy individuals whose biomechanical aging profiles resemble those of MCI or AD patients, suggesting they might be in a preclinical stage of neurodegeneration.
  • Localized Aging Patterns: The analysis reveals that aging effects are not uniform across the brain, with distinct regional contributions to age prediction that evolve throughout the lifespan.

Key Results

  • Improved Brain Age Prediction: The MRE-based models, particularly with the self-supervised deep learning approach, achieved significantly lower Mean Absolute Errors (MAE) in brain age prediction compared to MRI-based models. For instance, self-supervised MRE models achieved an MAE of 3.51 years, a substantial improvement over MRI models.
  • Disease Detection: MRE-based models showed a clear differentiation in brain age gap (BAG) between healthy individuals and those with MCI or AD, whereas MRI models showed minimal differences.
  • For MCI, damping ratio was the primary driver of BAG elevation.
  • For AD, stiffness was the dominant contributor to BAG acceleration.
  • Regional Contributions: Stiffness-based predictions showed a shift from cortical to subcortical regions with age, while damping ratio-based predictions showed a decline in cortical relevance in older adults. The thalamus and caudate consistently emerged as important regions.

Limitations and Future Work

  • Dataset Size: While large for MRE studies, the dataset is smaller than typical MRI datasets, which might explain the lack of statistical significance in MRI-based models for AD detection. However, this also highlights MRE's increased sensitivity.
  • Inter-Study Variability: Although mitigated by image registration, some inter-study variability might persist due to the aggregation of data from multiple clinical studies.
  • Retraining Models: The need to retrain models without healthy controls for disease cohorts makes them susceptible to dataset biases.
  • Future Directions:
    • Longitudinal Studies: Future research should conduct large-scale longitudinal studies to prospectively identify individuals at risk for cognitive impairment and dementia.
  • Multimodal Integration: Integrating MRE with other advanced MRI-based microstructural imaging techniques and genetic/lifestyle factors could further enhance sensitivity and clinical applicability.
  • Clinical Applicability: Further research is needed to explore the clinical utility of these MRE-based biomarkers for early detection and intervention.
  • Hardware/Sequence Advancements: Continued advancements in MRE hardware and sequence design could help mitigate challenges like partial brain coverage and brainstem measurement difficulties.

Summary updated

Abstract

Abstract The aging process affects brain structure and function, yet its biomechanical properties remain underexplored. Magnetic Resonance Elastography (MRE) provides a unique perspective by mapping brain tissue stiffness and damping ratio, observables that correlate with age and disease. Using a self-supervised contrastive regression framework, we demonstrate that MRE surpasses conventional structural magnetic resonance imaging (MRI) in sensitivity. Specifically, stiffness captures Alzheimer’s disease (AD), while damping ratio detects subtle changes associated with mild cognitive impairment (MCI). Our regional analysis identifies deep brain structures, particularly the caudate and thalamus, as key biomarkers of aging. The greater age sensitivity of MRE translates to superior differentiation of AD and MCI from healthy individuals, pinpointing regions where significant biomechanical alterations occur, notably the thalamus in AD and hippocampus in MCI. Furthermore, our results reveal biomechanical alterations in cognitively healthy individuals whose aging profiles closely resemble patients with MCI and AD. These findings highlight MRE’s potential as a biomarker for early neurodegenerative changes, aiding dementia risk detection and early intervention.
BibTeX
@misc{trauble2025enhancing,
  author = {Jakob Träuble and Lucy V. Hiscox and Curtis L. Johnson and Angelica I. Avilés-Rivero and Carola‐Bibiane Schönlieb and Gabriele S. Kaminski Schierle},
  title = {{Enhancing Brain Age Prediction and Neurodegeneration Detection with Contrastive Learning on Regional Biomechanical Properties}},
  year = {2025},
  doi = {10.1101/2025.03.25.645330},
  howpublished = {bioRxiv},
}

Loading related content...

Figure