Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Enhancing Alzheimer's Diagnosis: Leveraging Anatomical Landmarks in Graph Convolutional Neural Networks on Tetrahedral Meshes

Yanxi Chen, Mohammad Farazi, Zhangsihao Yang, Yonghui Fan, Nicholas Ashton, Eric M Reiman, Yi Su, Yalin Wang

TL;DR

This paper tackles Alzheimer's disease diagnosis from noninvasive MRI by leveraging a geometry-aware transformer on tetrahedral meshes. It introduces LETetCNN, which uses Gaussian Process–based anatomical landmarks for patch-based tokenization and sparse local attention, enabling scalable analysis of large meshes while preserving global structure. The model fuses learned geometric features with blood-based biomarkers such as $pTau$-217, achieving state-of-the-art performance in AD classification (e.g., $91.7\%$ accuracy for AD vs CN) and brain amyloid positivity prediction (up to $0.862$ accuracy in the medium-risk group) with strong Grad-CAM evidence supporting clinically relevant regions. The approach reduces dependence on costly PET scans and demonstrates potential for integrating structural MRI with biomarker data to improve early and differential AD diagnosis, with scalability to large volumetric meshes and future extensions to multi-scale tokenization and richer anatomical priors.

Abstract

Alzheimer's disease (AD) is a major neurodegenerative condition that affects millions around the world. As one of the main biomarkers in the AD diagnosis procedure, brain amyloid positivity is typically identified by positron emission tomography (PET), which is costly and invasive. Brain structural magnetic resonance imaging (sMRI) may provide a safer and more convenient solution for the AD diagnosis. Recent advances in geometric deep learning have facilitated sMRI analysis and early diagnosis of AD. However, determining AD pathology, such as brain amyloid deposition, in preclinical stage remains challenging, as less significant morphological changes can be observed. As a result, few AD classification models are generalizable to the brain amyloid positivity classification task. Blood-based biomarkers (BBBMs), on the other hand, have recently achieved remarkable success in predicting brain amyloid positivity and identifying individuals with high risk of being brain amyloid positive. However, individuals in medium risk group still require gold standard tests such as Amyloid PET for further evaluation. Inspired by the recent success of transformer architectures, we propose a geometric deep learning model based on transformer that is both scalable and robust to variations in input volumetric mesh size. Our work introduced a novel tokenization scheme for tetrahedral meshes, incorporating anatomical landmarks generated by a pre-trained Gaussian process model. Our model achieved superior classification performance in AD classification task. In addition, we showed that the model was also generalizable to the brain amyloid positivity prediction with individuals in the medium risk class, where BM alone cannot achieve a clear classification. Our work may enrich geometric deep learning research and improve AD diagnosis accuracy without using expensive and invasive PET scans.

Enhancing Alzheimer's Diagnosis: Leveraging Anatomical Landmarks in Graph Convolutional Neural Networks on Tetrahedral Meshes

TL;DR

This paper tackles Alzheimer's disease diagnosis from noninvasive MRI by leveraging a geometry-aware transformer on tetrahedral meshes. It introduces LETetCNN, which uses Gaussian Process–based anatomical landmarks for patch-based tokenization and sparse local attention, enabling scalable analysis of large meshes while preserving global structure. The model fuses learned geometric features with blood-based biomarkers such as -217, achieving state-of-the-art performance in AD classification (e.g., accuracy for AD vs CN) and brain amyloid positivity prediction (up to accuracy in the medium-risk group) with strong Grad-CAM evidence supporting clinically relevant regions. The approach reduces dependence on costly PET scans and demonstrates potential for integrating structural MRI with biomarker data to improve early and differential AD diagnosis, with scalability to large volumetric meshes and future extensions to multi-scale tokenization and richer anatomical priors.

Abstract

Alzheimer's disease (AD) is a major neurodegenerative condition that affects millions around the world. As one of the main biomarkers in the AD diagnosis procedure, brain amyloid positivity is typically identified by positron emission tomography (PET), which is costly and invasive. Brain structural magnetic resonance imaging (sMRI) may provide a safer and more convenient solution for the AD diagnosis. Recent advances in geometric deep learning have facilitated sMRI analysis and early diagnosis of AD. However, determining AD pathology, such as brain amyloid deposition, in preclinical stage remains challenging, as less significant morphological changes can be observed. As a result, few AD classification models are generalizable to the brain amyloid positivity classification task. Blood-based biomarkers (BBBMs), on the other hand, have recently achieved remarkable success in predicting brain amyloid positivity and identifying individuals with high risk of being brain amyloid positive. However, individuals in medium risk group still require gold standard tests such as Amyloid PET for further evaluation. Inspired by the recent success of transformer architectures, we propose a geometric deep learning model based on transformer that is both scalable and robust to variations in input volumetric mesh size. Our work introduced a novel tokenization scheme for tetrahedral meshes, incorporating anatomical landmarks generated by a pre-trained Gaussian process model. Our model achieved superior classification performance in AD classification task. In addition, we showed that the model was also generalizable to the brain amyloid positivity prediction with individuals in the medium risk class, where BM alone cannot achieve a clear classification. Our work may enrich geometric deep learning research and improve AD diagnosis accuracy without using expensive and invasive PET scans.

Paper Structure

This paper contains 22 sections, 9 equations, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Notation
  4. Gaussian Process-based Anatomical Landmarking
  5. Tokenization
  6. Node and Patch-Level Embeddings
  7. Super Node Representation
  8. Radius Graph Construction
  9. Relative Positional Encoding
  10. Attention Mechanism
  11. Computational Efficiency
  12. Experimental Design
  13. Datasets
  14. Data Preprocessing
  15. Model Training
  16. ...and 7 more sections

Figures (3)

  • Figure 1: The architecture of our model begins with identifying patch centers using Gaussian Process (GP)-based landmarks. Local-aware features are then learned through TetCNN, followed by constructing a radius graph to compute sparse attention over the previously learned tokens. Additionally, we pre-compute the Laplace-Beltrami Operator (LBO) and landmarks to streamline the process.
  • Figure 2: Visualization of tokenization of the tetrahedral mesh. We assigned each node of the mesh the the nearest super node to create our tokens, resulting in patchwise visualization of the mesh. The supernodes are located in the centers of each patch. Each color represent unique patch (token).
  • Figure 3: Grad-CAM visualization highlighting regions of interest that align with Alzheimer's Disease (AD) pathology. Both results from AD classification task (A-C) and brain amyloid positivity prediction task (D-F) exhibited strong activations in the temporal gyrus and parts of the inferior temporal lobe, consistent with early cortical atrophy in AD. Mild activation was also observed in superior frontal cortex, which was related to later-stage disease progression. These results closely align with neuroimaging studies identifying key biomarkers for AD pathology.