Deep learning classifies craniofacial CT slices with 96.46% ensemble accuracy

[medRxiv] A preprint study reports deep learning models that classify craniofacial CT slices into six anatomical categories. A total of 11,769 images were collected from a multicenter database involving 10 Japanese university hospitals and the public Head-Neck-Radiomics-HN1 dataset; patients were divided into 70% training and 30% evaluation sets. Five CNN architectures were trained from scratch, and hard- and soft-voting ensembles were evaluated. Individual-model accuracy ranged from 94.72% to 96.09%, and the hard-voting ensemble reached 96.46% accuracy (Cohen kappa = 0.954). According to the authors, performance differed among individual models (Cochran Q = 99.29, p < 0.001), but the ensemble did not significantly outperform EfficientNetV2B1 (McNemar p = 0.163); errors occurred mainly between adjacent anatomical regions. The authors said deep learning can classify craniofacial CT slices accurately and may support automated dataset organization and AI-assisted fracture diagnosis, adding that external validation is needed.