Background: Early assessment of cranial deformities in newborns, such as plagiocephaly, brachycephaly, dolichocephaly, turricephaly, and trigonocephaly, requires precise and non-invasive methods. Methodology: This study presents a methodology using a 3D scanning smartphone application to capture three-dimensional head point clouds. A total of 60 3D point cloud cases were classified according to six classes of head deformities. These 60 real 3D point clouds were expanded to 3600 semi-synthetic point clouds via controlled geometric transformations simulating realistic cranial variations. A total of 138 morphometric descriptors were extracted per class, representing spatial head features as distances from the centre of the point cloud to the head surface. These descriptors were used to train and compare three machine learning models: decision tree, random forest, and multilayer perceptron, enabling the automatic classification of six infant’s head deformities. Results: The machine learning models achieved high classification accuracy, with F1-scores up to 0.98, demonstrating the effectiveness of the approach. Conclusions: The results demonstrate the potential of combining mobile 3D sensors, image-based modelling, semi-synthetic data, and artificial intelligence to provide predictive support in clinical applications, highlighting the usefulness of low-cost portable optical sensors.
Quispe-Enriquez et al. (2026) studied this question.