Craniofacial Growth Transformer: A Growth-Aware Framework for Longitudinal Craniofacial Analysis and Forecasting
Abstract Accurate characterization of craniofacial growth is important for orthodontic planning, craniofacial assessment, and long-term monitoring of facial development. Most existing methods rely on sparse or cross-sectional data. As a result, they do not fully capture how facial structures change over time or how different facial regions develop together.This study presents a growth-aware framework for longitudinal craniofacial analysis and forecasting. The framework is based on a dense longitudinal dataset containing nearly ten years of near-daily facial observations from three individuals. Using this dataset, we analyze temporal dependencies, age-related differences, developmental consistency across individuals, and relationships among craniofacial measurements.The results show that several craniofacial measurements follow similar developmental trajectories across individuals. This finding suggests the presence of shared growth patterns despite individual differences. The analysis also identifies stable relationships between anatomically related regions, including bilateral ocular structures and mandibular-facial dimensions. These findings indicate that craniofacial growth follows a coordinated process rather than independent changes in individual features.Based on these findings, we propose a Long Short-Term Memory–Temporal–Spatial Transformer (LSTM-TS-TF) model. The model captures temporal continuity, long-range developmental dependencies, and spatial interactions between facial measurements. Experimental results show that the proposed model improves forecasting accuracy and reduces prediction error by approximately 63--72% compared with conventional LSTM models.These findings show that dense longitudinal modeling can identify interpretable growth descriptors and coordinated developmental patterns. The results also support the use of growth-aware deep learning models for data-driven craniofacial analysis and predictive modeling in biomedical applications.
Authors
- Mostafa A. Salama (ORCID: https://orcid.org/0000-0003-2559-8056)
Institutions
- British University in Egypt (EG)
Publication Details
- Journal
- Research Square
- Published
- 2026-11-16
- DOI
- https://doi.org/10.21203/rs.3.rs-11010034/v1
- Primary Topic
- Orthodontics and Dentofacial Orthopedics
- Type
- preprint