Plain film mandibular fracture detection using machine learning – Model development

Abstract

The mandible is the second most fractured bone of the facial skeleton. The most common imaging modality for diagnosis are the Orthopantomogram (OPG) and posterior-anterior mandible (PAM) x-rays. This study was designed to develop a machine learning (ML) model for use to detect mandibular fractures on both OPG and PAM. 2000 consecutive incidences of orders for mandibular imaging were retrospectively collected, with 409 incidences of orders performed for the indication of trauma, and 117 incidences with fractures. These were used to train and validate the developed model. The best ML model achieved a precision of 81.9, recall of 71.9, mean Average Precision (mAP) @ 0.5 of 72.6 and F1-score of 76.5. Current research within the ML field on mandibular fractures is not standardised which makes it difficult to compare results across different datasets. ML in this research area requires standardisation and models require further development with heterogeneous and clinically relevant datasets to prove useful within the clinical environment.