A feasibility study of deep learning-based segmentation of the inferior alveolar nerve on magnetic resonance neurography

Abstract

The inferior alveolar nerve (IAN) is the major sensory nerve innervating the mandibular region, and its automatic segmentation is crucial. It has been indirectly identified on the mandibular canal, which surrounds the IAN, using computed tomography (CT) and cone-beam computed tomography (CBCT). Magnetic resonance neurography (MRN) is an imaging technique designed for nerve visualization, facilitating discrimination of small peripheral nerves from surrounding soft tissues. To our knowledge, this study is the first to perform semi-automatic segmentation of the IAN using MRN images. We developed a deep learning model based on 6,027 coronal MRN images and evaluated its performance using four quantitative metrics, comparing it with six state-of-the-art models that were retrained and tested on the same dataset for small-structure segmentation. Our model achieved a dice similarity coefficient (DSC) of 0.712 +/- 0.254, significantly outperforming the six comparator models. In addition, in an analysis of segmentation failure rates according to DSC thresholds, our model demonstrated the lowest failure rate. In conclusion, unlike many previous studies that focused on bony boundaries using CBCT or CT, this study demonstrates the feasibility and potential clinical utility of MRN-based IAN segmentation.