Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Distance Transform Guided Mixup for Alzheimer's Detection

Zobia Batool, Huseyin Ozkan, Erchan Aptoula

TL;DR

The paper tackles domain shift in Alzheimer's disease MRI classification by proposing Distance Transform Guided Mixup (DTGM), a structure-aware augmentation built on a 3D U-Net backbone. By computing an offline distance transform $D(x)$ and using region masks derived from thresholds to mix regions from pairs of scans, DTGM preserves brain structure while increasing data diversity; mixed labels are weighted by region-based pixel contributions and trained with weighted soft cross-entropy to handle class imbalance. The method demonstrates improved generalization across external datasets (ADNI and AIBL) compared to MixUp, RSC, and CCSDG baselines, achieving notable gains in accuracy and F1 under domain shift. This approach offers a practical path toward more robust, cross-site Alzheimer's detection and can be extended to other neurodegenerative tasks facing similar domain variations.

Abstract

Alzheimer's detection efforts aim to develop accurate models for early disease diagnosis. Significant advances have been achieved with convolutional neural networks and vision transformer based approaches. However, medical datasets suffer heavily from class imbalance, variations in imaging protocols, and limited dataset diversity, which hinder model generalization. To overcome these challenges, this study focuses on single-domain generalization by extending the well-known mixup method. The key idea is to compute the distance transform of MRI scans, separate them spatially into multiple layers and then combine layers stemming from distinct samples to produce augmented images. The proposed approach generates diverse data while preserving the brain's structure. Experimental results show generalization performance improvement across both ADNI and AIBL datasets.

Distance Transform Guided Mixup for Alzheimer's Detection

TL;DR

The paper tackles domain shift in Alzheimer's disease MRI classification by proposing Distance Transform Guided Mixup (DTGM), a structure-aware augmentation built on a 3D U-Net backbone. By computing an offline distance transform and using region masks derived from thresholds to mix regions from pairs of scans, DTGM preserves brain structure while increasing data diversity; mixed labels are weighted by region-based pixel contributions and trained with weighted soft cross-entropy to handle class imbalance. The method demonstrates improved generalization across external datasets (ADNI and AIBL) compared to MixUp, RSC, and CCSDG baselines, achieving notable gains in accuracy and F1 under domain shift. This approach offers a practical path toward more robust, cross-site Alzheimer's detection and can be extended to other neurodegenerative tasks facing similar domain variations.

Abstract

Alzheimer's detection efforts aim to develop accurate models for early disease diagnosis. Significant advances have been achieved with convolutional neural networks and vision transformer based approaches. However, medical datasets suffer heavily from class imbalance, variations in imaging protocols, and limited dataset diversity, which hinder model generalization. To overcome these challenges, this study focuses on single-domain generalization by extending the well-known mixup method. The key idea is to compute the distance transform of MRI scans, separate them spatially into multiple layers and then combine layers stemming from distinct samples to produce augmented images. The proposed approach generates diverse data while preserving the brain's structure. Experimental results show generalization performance improvement across both ADNI and AIBL datasets.

Paper Structure

This paper contains 10 sections, 11 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Model Architecture
  4. Distance Transform-Based Mixup Augmentation
  5. Label Mixing
  6. Experiments
  7. Datasets
  8. Experimental Settings
  9. Results and Discussion
  10. Conclusion

Figures (2)

  • Figure 1: Overview of the proposed Alzheimer's disease classification pipeline. 3D MRI scans are preprocessed before applying region mixing augmentation. The augmented images are then processed through a U-Net 3D architecture followed by the classifier.
  • Figure 2: Overview of the mixing strategy. Given two input MRI scans ($x_a$ and $x_b$), region-wise masks ($R_1, R_2, R_3, R_4$) are extracted to generate mixed samples ($\text{Mixed\_x}_{_a}$ and $\text{Mixed\_x}_{_b}$).