Implicit Neural Representations for Brain Aging Trajectory Classification

Implementation of our paper "Semi-Disentangled Spatiotemporal Implicit Neural Representations of Longitudinal Neuroimaging Data for Trajectory Classification," accepted at the LMID workshop at MICCAI 2025.

[Paper] [Web Page - TBA] [Corresponding Author Email]

---

Abstract

The human brain undergoes dynamic, potentially pathology-driven, structural changes throughout a lifespan. Longitudinal Magnetic Resonance Imaging (MRI) and other neuroimaging data are valuable for characterizing trajectories of change associated with typical and atypical aging. However, the analysis of such data is highly challenging given their discrete nature with different spatial and temporal image sampling patterns within individuals and across populations. This leads to computational problems for most traditional deep learning methods that cannot represent the underlying continuous biological process. To address these limitations, we present a new, fully data-driven method for representing aging trajectories across the entire brain by modelling subject-specific longitudinal T1-weighted MRI data as continuous functions using implicit neural representations (INRs).

The main contributions of this work are as follows:

1. We propose an INR architecture with semi-disentangled spatial and temporal parameters to model subject-level aging trajectories across the entire brain from longitudinal MRI data. We show that these INRs can be flexibly adapted to irregular temporal sampling.

2. We propose a method for directly classifying the brain aging trajectories encoded by INRs.

3. We develop a brain aging simulation framework to generate realistic, longitudinal 3D imaging data of healthy and Alzheimer's-like trajectories. We use this simulation to validate our methods.

---

Setup

This code was developed and tested using Python 3.12.7 and CUDA 12.4. To replicate the conda environment used in this work, run the following after cloning our repo:

conda env create -f environment.yml

Data

Our brain aging trajectory simulation was tested on longitudinal data produced by our 3D Conditional Diffusion Model. To reproduce the data used in all experiments:

TBA

To create data splits for subjects with regular and irregular sampling:

preprocess.py --data-dir <path/to/data-dir>

Experiments

We provide shell scripts with all necessary hyperparameters used for training INRs and classifying their brain aging trajectories:

├── scripts
│   ├── inr_init.sh
│   ├── inr_adapt.sh
│   ├── classify_inr.sh
│   └── classify_baseline.sh

---

Citation

Please cite our paper if you find this work useful:

{TBA}

Acknowledgements

We gratefully acknowledge this code is based on and incorporates other publically available projects, including:

• Neonatal Development INR from Bieder et al., 2024.
• inr2vec from De Luigi, Cardace, et al., 2023.
• WIRE from Saragadam et al., 2023.
• Implicit Segmentation from Stolt-Ansó et al., 2023.