From PDF Records to Dental View Classification with a Human-in-the-Loop Imaging Pipeline

Oral health is increasingly recognized as a public priority due to its strong association with overall well-being. However, both evaluator training and computational model development are constrained by the scarcity of publicly available, context-rich dental image datasets. This study presents a reproducible data-engineering pipeline, named SBBrasil TrainSheets, that transforms public PDF manuals from national oral health programs into structured, analyzable image corpora suitable for translational research. The proposed system automates image extraction, organizes volunteer-captured anatomical views, and embeds features for unsupervised quality control. A human-in-the-loop labeling interface prioritizes uncertain samples to improve annotation efficiency. As a proof-of-concept, a lightweight baseline classifier was trained to distinguish dental views (frontal vs. occlusal) using a volunteer-level split, achieving 96.4% accuracy after a short calibration phase. Overall, the pipeline enables reproducible dataset construction from existing public materials and supports quality assurance, educational applications, and downstream computational analysis in dental public health.