Quantification of Craniofacial Growth Pattern Based on Deep Learning

Overview

This study developed an end-to-end deep learning framework to quantify craniofacial growth patterns and sexual dimorphism from lateral cephalometric radiographs without requiring manual landmarking. The model was trained on a large dataset of 41,625 individuals aged 4–18 years to autonomously extract age-related and sex-related imaging features, addressing limitations of traditional cephalometric analysis which relies on subjective manual annotation and oversimplifies complex morphological changes.

Methods and approach

The research employed a deep learning architecture applied to lateral cephalometric radiographs from 41,625 pediatric and adolescent subjects spanning ages 4–18 years. The model was designed to operate without manual annotations, autonomously learning dynamic imaging features associated with continuous age intervals and sexual dimorphism. Gradient-weighted Class Activation Mapping was utilized to generate population-averaged saliency maps highlighting age-related and sex-related patterns of significance. Two novel quantitative indices were introduced: the Age-related Saliency Index to prioritize developmental importance across craniofacial regions, and the Sex-related Saliency Index to quantify the evolution of sexual dimorphic characteristics throughout development.

Key Findings

Age-related saliency maps revealed that developmental significance extended beyond external bone contours to internal anatomical details, with the Age-related Saliency Index providing quantitative prioritization of regions throughout the craniofacial skeleton during growth. The Sex-related Saliency Index demonstrated that sexually dimorphic characteristics were widely distributed across cranial bones in early developmental stages but progressively concentrated in the mandibular region by adulthood. These spatiotemporal patterns were visualized through population-averaged saliency maps that objectively identified regions of developmental and dimorphic importance.

Implications

The framework establishes an objective, annotation-free approach to analyzing large-scale cephalometric datasets, providing clinicians with quantitative references for assessing developmental stages and determining optimal intervention timing for specific craniofacial regions. By eliminating manual landmarking dependency, the model improves generalizability and robustness while accounting for multi-tissue complexity inherent in craniofacial morphology. The generated saliency maps and quantitative indices offer both validation of established developmental theories and novel insights into coordinated skeletal growth patterns.

The identification of sex-specific radiological characteristics throughout development supports clinical decision-making in orthodontics and maxillofacial surgery by providing objective population-level references for normal developmental variation. The spatiotemporal dynamics revealed through this analysis clarify how sexual dimorphism in craniofacial structure emerges and evolves from early childhood through adulthood, with implications for understanding the regional specificity of sex-dependent morphological changes. These objective metrics may facilitate standardization of radiological assessment protocols across clinical settings.

Scope and limitations

This summary is based on the study abstract and available metadata. It does not include a full analysis of the complete paper, supplementary materials, or underlying datasets unless explicitly stated. Findings should be interpreted in the context of the original publication.