MOLAR: MRI-based Opportunistic Localization and Recognition of teeth

MOLAR is an open pipeline for analyzing teeth from routine brain MRI scans without requiring additional scanning. The research showed that tooth loss is associated with cognitive decline, yet dental status information is rarely available in neuroimaging studies. The team trained neural networks on 1,261 annotated MRI scans with 63,829 tooth markers. The automated model achieved F1 precision of 0.873 in tooth detection and counted teeth with a mean absolute error of 1.96 teeth. In an independent cohort of 233 patients, automated tooth counting reproduced the manual association with the Montreal Cognitive Assessment test (Pearson r = +0.372 versus +0.414). The software will be released as open source.