Geometric-Guided Few-Shot Dental Landmark Detection with Human-Centric Foundation Model

2025-07-07

AI Summary

To address the scarcity of anatomical landmark annotations and high expert annotation costs in dental cone-beam computed tomography (CBCT), leading to suboptimal detection accuracy, this paper proposes GeoSapiens: a geometrically aware framework built upon the Sapiens vision foundation model. It employs transfer learning with a novel geometric-guided loss function that explicitly encodes inter-tooth spatial and topological constraints, integrated with a few-shot fine-tuning strategy. Evaluated on an anterior teeth CBCT dataset, GeoSapiens achieves a 92.3% success detection rate under a stringent 0.5 mm error threshold—surpassing the state-of-the-art by 8.18 percentage points. The method significantly enhances automation and clinical robustness for orthodontic, periodontal, and pre-implantation assessments. Moreover, it establishes a generalizable, geometry-aware paradigm for keypoint localization in medical imaging under limited-label settings.

Technology Category

• Computer Vision: Remote Sensing / Geospatial AI
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Constraint Satisfaction and Optimization: Satisfiability Modulo Theories

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics

Abstract

Accurate detection of anatomic landmarks is essential for assessing alveolar bone and root conditions, thereby optimizing clinical outcomes in orthodontics, periodontics, and implant dentistry. Manual annotation of landmarks on cone-beam computed tomography (CBCT) by dentists is time-consuming, labor-intensive, and subject to inter-observer variability. Deep learning-based automated methods present a promising approach to streamline this process efficiently. However, the scarcity of training data and the high cost of expert annotations hinder the adoption of conventional deep learning techniques. To overcome these challenges, we introduce GeoSapiens, a novel few-shot learning framework designed for robust dental landmark detection using limited annotated CBCT of anterior teeth. Our GeoSapiens framework comprises two key components: (1) a robust baseline adapted from Sapiens, a foundational model that has achieved state-of-the-art performance in human-centric vision tasks, and (2) a novel geometric loss function that improves the model's capacity to capture critical geometric relationships among anatomical structures. Experiments conducted on our collected dataset of anterior teeth landmarks revealed that GeoSapiens surpassed existing landmark detection methods, outperforming the leading approach by an 8.18% higher success detection rate at a strict 0.5 mm threshold-a standard widely recognized in dental diagnostics. Code is available at: https://github.com/xmed-lab/GeoSapiens.

Problem

• Automates dental landmark detection to reduce manual annotation effort
• Addresses data scarcity in training deep learning models for dentistry
• Improves geometric accuracy in detecting anatomical structures in CBCT scans

Innovation

• Few-shot learning framework for dental landmarks
• Human-centric foundation model adaptation
• Geometric loss enhances anatomical relationships