Image-based mandibular and maxillary parcellation and annotation using computed tomography (IMPACT): a deep...

Accurate delineation of orodental structures on radiotherapy computed tomography (CT) images is essential for dosimetric assessment and dental decisions. We propose a deep-learning (DL) auto-segmentation framework for individual teeth and mandible/maxilla sub-volumes aligned with the ClinRad osteoradionecrosis staging system.

Mandible and maxilla sub-volumes were manually defined on simulation CT images from 60 clinical cases, differentiating alveolar from basal regions; teeth were labelled individually. For each task, a DL segmentation model was independently trained. A Swin UNETR-based model was used for mandible sub-volumes. For smaller structures (e.g., teeth and maxilla sub-volumes) a two-stage model first used the ResUNet to segment the entire teeth and maxilla regions as a single ROI used to crop the image input for Swin UNETR. In addition to segmentation accuracy and geometric precision, a dose-volume comparison was made between manual and model-predicted segmentations.

Segmentation performance varied across sub-volumes - mean Dice values of 0.85 (mandible basal), 0.82 (mandible alveolar), 0.78 (maxilla alveolar), 0.80 (upper central teeth), 0.69 (upper premolars), 0.76 (upper molars), 0.76 (lower central teeth), 0.70 (lower premolars), 0.71 (lower molars) - with limited applicability in segmenting sub-volumes absent in the data. The maxilla alveolar central sub-volume showed a statistically significant dose-volume difference in both D_mean and D_2%.

We present a novel DL-based auto-segmentation framework of orodental structures, enabling spatial localization of dose-related differences. This tool may enhance image-based bone injury detection and improve clinical decision-making in radiation oncology and dental care for head and neck cancer patients.