From notes to neurons: An open-source AI pipeline for 3D laryngeal modeling in anatomy and music education

The larynx is a complex anatomical structure central to human voice production, playing a critical role in clinical diagnosis, surgical planning, and rehabilitation. This study aims to develop an open-source, artificial intelligence-based three-dimensional (3D) laryngeal modeling pipeline using medical imaging data. Our approach integrates deep learning-based automatic segmentation (3D U-Net architecture), surface extraction (Marching Cubes), and multi-stage mesh optimization (Laplacian/Taubin smoothing, polygon decimation) within a unified framework, enabling high-accuracy modeling of the larynx's complex anatomy. Segmentation performance achieved a Dice coefficient of 0.87 for vocal folds and 0.90 for the thyroid cartilage, demonstrating comparability with similar studies in the literature. Following 3D reconstruction, mesh optimization reduced the average polygon count by 42% while preserving anatomical integrity. All software components are shared as open source, and the system is designed with a modular architecture adaptable to different anatomical regions. This study provides a reproducible and accessible modeling pipeline in the field of computational anatomy.