Three-Dimensional Radiographic and Morphological Features of Ameloblastoma and Odontogenic Keratocyst: A Comparative Cone Beam Computed Tomography Study Using Automatic Segmentation.

OBJECTIVES: We aimed to compare the three-dimensional (3D) radiographic and morphological features, and explore their associations in ameloblastoma (AM) and odontogenic keratocyst (OKC) on CBCT scans, to facilitate clinicians in the preoperative differential diagnosis between the two lesions. METHODS: CBCT scans of 151 AMs and 164 OKCs were retrospectively collected. The morphological features (volume, length, width, height) of lesions were obtained by OCL-Net automatically, an algorithm established for odontogenic cystic lesion (OCL) segmentation. Meanwhile, the radiographic features of lesions including shape, border, buccolingual expansion, cortical bone destruction, lesion-tooth relationship, root resorption and tooth displacement were assessed by maxillofacial radiologists. Differences in morphological and radiographic features between AM and OKC were identified. Additionally, associations between morphological and radiographic features were analyzed separately in AM and OKC cases. RESULTS: AM showed significantly greater volume (p = 0.002), length (p = 0.009), length-to-width (p = 0.014), and length-to-height (p = 0.022) ratios compared with OKC. Additionally, scalloped borders (p = 0.003), buccolingual expansion (p < 0.001), cortical bone destruction (p = 0.035), root resorption (p < 0.001) and tooth displacement (p = 0.014), were more commonly observed in AMs than in OKCs. Moreover, OKCs with cortical bone destruction (p < 0.001) and tooth displacement (p < 0.01) were significantly larger than those without, while these radiographic features were not associated with lesion volume in AMs. CONCLUSION: OKCs generally show smaller lesion volumes, less extensive invasion along the frontal axis, and less aggressive radiographic features on CBCT than AMs. In addition, small lesions that cause destruction of the surrounding tissues are more likely to be AMs rather than OKCs. CLINICAL RELEVANCE: Our findings may aid clinicians in the preoperative differentiation of AM and OKC, while our proposed OCL-Net may facilitate the future development of a fully automated diagnostic system and provide a novel perspective for the differential diagnosis of a broader spectrum of jaw space-occupying lesions.