Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ADAPT: Alzheimer Diagnosis through Adaptive Profiling Transformers

Yifeng Wang, Ke Chen, Haohan Wang

TL;DR

This work addresses the challenge of diagnosing Alzheimer’s disease from 3D MRI while mitigating the high computational cost of 3D models. It introduces ADAPT, a pure 2D Vision Transformer framework that decomposes 3D MRIs into three orthogonal 2D slice sequences and uses a staged set of encoders (SAE, DS-AE, IntraCAE, InterCAE) with a fusion attention mechanism and morphology augmentation. An adaptive training strategy guides the model to focus on the most informative dimensions, enabling efficient learning with fewer parameters. Experiments on ADNI and cross-domain datasets show ADAPT achieves state-of-the-art accuracy with strong generalization and reduced memory usage, with attention maps aligning to known AD-affected brain regions.

Abstract

Automated diagnosis of Alzheimer Disease(AD) from brain imaging, such as magnetic resonance imaging (MRI), has become increasingly important and has attracted the community to contribute many deep learning methods. However, many of these methods are facing a trade-off that 3D models tend to be complicated while 2D models cannot capture the full 3D intricacies from the data. In this paper, we introduce a new model structure for diagnosing AD, and it can complete with performances of 3D models while essentially is a 2D method (thus computationally efficient). While the core idea lies in new perspective of cutting the 3D images into multiple 2D slices from three dimensions, we introduce multiple components that can further benefit the model in this new perspective, including adaptively selecting the number of sclices in each dimension, and the new attention mechanism. In addition, we also introduce a morphology augmentation, which also barely introduces new computational loads, but can help improve the diagnosis performances due to its alignment to the pathology of AD. We name our method ADAPT, which stands for Alzheimer Diagnosis through Adaptive Profiling Transformers. We test our model from a practical perspective (the testing domains do not appear in the training one): the diagnosis accuracy favors our ADAPT, while ADAPT uses less parameters than most 3D models use.

ADAPT: Alzheimer Diagnosis through Adaptive Profiling Transformers

TL;DR

This work addresses the challenge of diagnosing Alzheimer’s disease from 3D MRI while mitigating the high computational cost of 3D models. It introduces ADAPT, a pure 2D Vision Transformer framework that decomposes 3D MRIs into three orthogonal 2D slice sequences and uses a staged set of encoders (SAE, DS-AE, IntraCAE, InterCAE) with a fusion attention mechanism and morphology augmentation. An adaptive training strategy guides the model to focus on the most informative dimensions, enabling efficient learning with fewer parameters. Experiments on ADNI and cross-domain datasets show ADAPT achieves state-of-the-art accuracy with strong generalization and reduced memory usage, with attention maps aligning to known AD-affected brain regions.

Abstract

Automated diagnosis of Alzheimer Disease(AD) from brain imaging, such as magnetic resonance imaging (MRI), has become increasingly important and has attracted the community to contribute many deep learning methods. However, many of these methods are facing a trade-off that 3D models tend to be complicated while 2D models cannot capture the full 3D intricacies from the data. In this paper, we introduce a new model structure for diagnosing AD, and it can complete with performances of 3D models while essentially is a 2D method (thus computationally efficient). While the core idea lies in new perspective of cutting the 3D images into multiple 2D slices from three dimensions, we introduce multiple components that can further benefit the model in this new perspective, including adaptively selecting the number of sclices in each dimension, and the new attention mechanism. In addition, we also introduce a morphology augmentation, which also barely introduces new computational loads, but can help improve the diagnosis performances due to its alignment to the pathology of AD. We name our method ADAPT, which stands for Alzheimer Diagnosis through Adaptive Profiling Transformers. We test our model from a practical perspective (the testing domains do not appear in the training one): the diagnosis accuracy favors our ADAPT, while ADAPT uses less parameters than most 3D models use.
Paper Structure (16 sections, 16 equations, 5 figures, 4 tables, 1 algorithm)

This paper contains 16 sections, 16 equations, 5 figures, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. 3D Vision Transformer
  4. Deep Learning for Medical Image Analysis
  5. Methodology
  6. ADAPT Architecture
  7. Fusion Attention Mechanism
  8. Morphology Augmentation
  9. Adaptive Training Strategy
  10. Experiments
  11. Experimental Dataset
  12. Implementation Details
  13. Evaluation Between Baselines
  14. Ablation Study
  15. Visualization Result
  16. ...and 1 more sections

Figures (5)

  • Figure 1: The overall framework of our proposed 2D ViT-Based Model for 3D Medical Imaging ( ADAPT).
  • Figure 2: The detailed architecture for our ADAPT. ADAPT consists of four main components: Self-Attention Encoders (SAE) across three views, Dimension-specific Self-Attention Encoders (DS-AE), Intra-dimension Cross-Attention Encoders (IntraCAE), Inter-dimension Cross-Attention Encoders (InterCAE). The figure shows more details in sagittal view for illustration purposes. In practice, the model will adaptively attend to different views.
  • Figure 3: The visualization of Alzheimer's Disease (AD) image, Normal Control (NC) image and Mild Cognitive Impairment (MCI) image. The left is the raw image and the right is the augmented image. The cerebral ventricle (red circle) has a significant difference in size for AD and NC.
  • Figure 4: Attention map for Normal Control result. Each line corresponds to one view dimension: saggital, coronal and axial.
  • Figure 5: Attention map for Alzheimer's Disease result. Each line corresponds to one view dimension: saggital, coronal and axial.