Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

3D Graph Attention Networks for High Fidelity Pediatric Glioma Segmentation

Harish Thangaraj, Diya Katariya, Eshaan Joshi, Sangeetha N

TL;DR

The paper tackles automated segmentation of pediatric gliomas, a challenging task due to infiltrative tumor growth, by introducing a 3D U-Net variant augmented with a Graph Cross Attention mechanism within a MedNeXt-based backbone. The method captures multi-scale features and focuses attention on tumor-relevant regions, achieving a Dice Similarity Coefficient of approximately 0.794 and an HD95 around 12 mm in cross-validated experiments, while maintaining efficient resource usage. Through a comprehensive preprocessing pipeline on the BraTS pediatric dataset and a multi-stage Graph Attention module, the approach demonstrates robust boundary precision and improved delineation of complex tumor structures. The work advances automated pediatric glioma segmentation with potential clinical impact by enabling more accurate, efficient tumor delineation for diagnosis and treatment planning.

Abstract

Pediatric brain tumors, particularly gliomas, represent a significant cause of cancer related mortality in children with complex infiltrative growth patterns that complicate treatment. Early, accurate segmentation of these tumors in neuroimaging data is crucial for effective diagnosis and intervention planning. This study presents a novel 3D UNet architecture with a spatial attention mechanism tailored for automated segmentation of pediatric gliomas. Using the BraTS pediatric glioma dataset with multiparametric MRI data, the proposed model captures multi-scale features and selectively attends to tumor relevant regions, enhancing segmentation precision and reducing interference from surrounding tissue. The model's performance is quantitatively evaluated using the Dice similarity coefficient and HD95, demonstrating improved delineation of complex glioma structured. This approach offers a promising advancement in automating pediatric glioma segmentation, with the potential to improve clinical decision making and outcomes.

3D Graph Attention Networks for High Fidelity Pediatric Glioma Segmentation

TL;DR

The paper tackles automated segmentation of pediatric gliomas, a challenging task due to infiltrative tumor growth, by introducing a 3D U-Net variant augmented with a Graph Cross Attention mechanism within a MedNeXt-based backbone. The method captures multi-scale features and focuses attention on tumor-relevant regions, achieving a Dice Similarity Coefficient of approximately 0.794 and an HD95 around 12 mm in cross-validated experiments, while maintaining efficient resource usage. Through a comprehensive preprocessing pipeline on the BraTS pediatric dataset and a multi-stage Graph Attention module, the approach demonstrates robust boundary precision and improved delineation of complex tumor structures. The work advances automated pediatric glioma segmentation with potential clinical impact by enabling more accurate, efficient tumor delineation for diagnosis and treatment planning.

Abstract

Pediatric brain tumors, particularly gliomas, represent a significant cause of cancer related mortality in children with complex infiltrative growth patterns that complicate treatment. Early, accurate segmentation of these tumors in neuroimaging data is crucial for effective diagnosis and intervention planning. This study presents a novel 3D UNet architecture with a spatial attention mechanism tailored for automated segmentation of pediatric gliomas. Using the BraTS pediatric glioma dataset with multiparametric MRI data, the proposed model captures multi-scale features and selectively attends to tumor relevant regions, enhancing segmentation precision and reducing interference from surrounding tissue. The model's performance is quantitatively evaluated using the Dice similarity coefficient and HD95, demonstrating improved delineation of complex glioma structured. This approach offers a promising advancement in automating pediatric glioma segmentation, with the potential to improve clinical decision making and outcomes.

Paper Structure

This paper contains 27 sections, 12 equations, 4 figures, 1 table, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Works
  3. Advancements in Brain Tumor Segmentation Techniques
  4. Dataset and Preprocessing
  5. BraTs Dataset
  6. Dataset Preparation and Preprocessing
  7. Standardization and Cropping
  8. Data Augmentation
  9. Dataset Splitting
  10. Binary Mask Creation
  11. Validation of Preprocessing
  12. Conclusion
  13. Proposed Methodology
  14. U-Net Model
  15. MedNeXt Architecture
  16. ...and 12 more sections

Figures (4)

  • Figure 1: Overview of the BraTs Dataset
  • Figure 2: Overview of the MedNext architecture for 3D medical imaging.
  • Figure 3: Five-fold cross-validation accuracies for the proposed architecture, averaged and for each tumor sub-region.
  • Figure 4: Five-fold cross-validation training and validation loss trends for the proposed model.