Attention Maps in 3D Shape Classification for Dental Stage Estimation with Class Node Graph Attention Networks

2025-09-09

AI Summary

Accurate Demirjian staging of third molars from CBCT-derived 3D dental meshes is critical in forensic age estimation and clinical risk assessment, yet deep learning models often lack interpretability and trustworthiness in high-stakes settings. Method: We propose the Class Node Graph Attention Network (CGAT), a graph-based architecture incorporating a global [CLS] node and directed attention edges. It fuses geometric features—local mean curvature and centroid distance—and employs graph attention convolution for end-to-end classification with built-in decision visualization. Attention heatmaps enable retrospective verification of morphologically salient regions by domain experts. Contribution/Results: Evaluated on real-world dental CBCT data, CGAT achieves a weighted F1-score of 0.76—significantly outperforming baseline models—while simultaneously optimizing both predictive performance and model interpretability. This work establishes a verifiable, traceable paradigm for deploying 3D medical image AI in clinical practice.

Technology Category

• Computer Vision: 3D Computer Vision
• Machine Learning: Graph-based Machine Learning
• Reasoning under Uncertainty: Graphical Models

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Deep learning offers a promising avenue for automating many recognition tasks in fields such as medicine and forensics. However, the black-box nature of these models hinders their adoption in high-stakes applications where trust and accountability are required. For 3D shape recognition tasks in particular, this paper introduces the Class Node Graph Attention Network (CGAT) architecture to address this need. Applied to 3D meshes of third molars derived from CBCT images, for Demirjian stage allocation, CGAT utilizes graph attention convolutions and an inherent attention mechanism, visualized via attention rollout, to explain its decision-making process. We evaluated the local mean curvature and distance to centroid node features, both individually and in combination, as well as model depth, finding that models incorporating directed edges to a global CLS node produced more intuitive attention maps, while also yielding desirable classification performance. We analyzed the attention-based explanations of the models, and their predictive performances to propose optimal settings for the CGAT. The combination of local mean curvature and distance to centroid as node features yielded a slight performance increase with 0.76 weighted F1 score, and more comprehensive attention visualizations. The CGAT architecture's ability to generate human-understandable attention maps can enhance trust and facilitate expert validation of model decisions. While demonstrated on dental data, CGAT is broadly applicable to graph-based classification and regression tasks, promoting wider adoption of transparent and competitive deep learning models in high-stakes environments.

Problem

• Explaining 3D shape classification decisions for dental stage estimation
• Enhancing trust in deep learning models via interpretable attention mechanisms
• Optimizing graph attention networks for transparent high-stakes medical applications

Innovation

• CGAT uses graph attention convolutions for explainable decisions
• Combines local curvature and centroid distance as node features
• Generates intuitive attention maps via attention rollout visualization