Search papers, labs, and topics across Lattice.
The FaceMesh2HPO framework utilizes a hierarchical PointNet-based pipeline to classify facial phenotypic descriptors aligned with the Human Phenotype Ontology, leveraging 3D facial meshes derived from 2D images annotated by 124 clinicians across 10 disorders. Key findings reveal that models integrating 3D facial geometry, facial outlines, and demographic metadata achieved AUROCs ranging from ~0.55 to ~0.89, with parent nodes outperforming leaf terms, indicating a nuanced understanding of phenotypic classification. However, the study highlights limitations in generalizability for rare disorders, suggesting a need for improved data diversity and feature selection to enhance clinical applicability.
Hierarchical modeling of 3D facial geometry reveals significant insights into phenotype classification, but struggles with rare conditions highlight critical gaps in current methodologies.
FaceMesh2HPO is a framework for classifying facial phenotypic descriptors aligned with the Human Phenotype Ontology (HPO) to support clinical diagnosis. Using annotations from 124 clinicians across 10 disorders (107 HPO terms) combined with non-syndromic controls, we generated 3D facial meshes (478 landmarks) from 2D images and trained a hierarchical PointNet-based pipeline with cascading classification and feature elimination. The best models, incorporating 3D meshes, facial outline, and demographic metadata, achieved AUROCs between ~0.55 and ~0.89, with higher performance at parent nodes than leaf terms. External validation showed variable generalizability across disorders. Results demonstrate that hierarchical modeling of 3D facial geometry enables interpretable, ontology-linked phenotype classification, though performance on rare leaf terms remains limited. Improved data diversity and feature selection strategies are needed to enhance robustness and clinical utility.
