Automated Estimation of Anatomical Risk Metrics for Endoscopic Sinus Surgery Using Deep Learning

2025-11-10

AI Summary

Endoscopic sinus surgery requires precise preoperative assessment of skull base anatomical risks—such as the Keros, Gera, and Thailand–Malaysia–Singapore (TMS) classifications—yet conventional manual measurements on coronal CT or CBCT are time-consuming and prone to inter-observer variability. To address this, we propose an end-to-end deep learning framework employing a global-to-local heatmap regression strategy to automatically localize key anatomical landmarks and directly estimate all three internationally standardized risk scores. This work represents the first unified automated method capable of concurrently computing multiple scoring systems. Quantitative evaluation shows mean absolute errors of 0.506 mm (Keros depth), 4.516° (Gera angle), and 0.777–0.802 mm (TMS classification metrics). The framework significantly improves landmark localization accuracy and measurement efficiency while reducing human-induced error, demonstrating strong potential for clinical deployment.

Technology Category

• Machine Learning: Neuro-Symbolic Learning
• Search and Optimization: Local Search
• Intelligent Robots: Localization, Mapping, and Navigation

Application Category

• Security and Privacy: Large-scale security measurements
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Abstract

Endoscopic sinus surgery requires careful preoperative assessment of the skull base anatomy to minimize risks such as cerebrospinal fluid leakage. Anatomical risk scores like the Keros, Gera and Thailand-Malaysia-Singapore score offer a standardized approach but require time-consuming manual measurements on coronal CT or CBCT scans. We propose an automated deep learning pipeline that estimates these risk scores by localizing key anatomical landmarks via heatmap regression. We compare a direct approach to a specialized global-to-local learning strategy and find mean absolute errors on the relevant anatomical measurements of 0.506mm for the Keros, 4.516{deg} for the Gera and 0.802mm / 0.777mm for the TMS classification.

Problem

• Automating anatomical risk score estimation for endoscopic sinus surgery
• Eliminating manual CT measurement requirements through deep learning
• Reducing surgical risks by localizing key anatomical landmarks automatically

Innovation

• Automated deep learning pipeline for anatomical risk scores
• Localizing key landmarks via heatmap regression method
• Global-to-local learning strategy reduces measurement errors