Automatic multi-anatomical skull structure segmentation of cone-beam computed tomography scans using 3D UNETR

Abstract

The segmentation of medical and dental images is a fundamental step in automated clinical decision support systems. It supports the entire clinical workflow from diagnosis, therapy planning, intervention, and follow-up. In this paper, we propose a novel tool to accurately process a full-face segmentation in about 5 minutes that would otherwise require an average of 7h of manual work by experienced clinicians. This work focuses on the integration of the state-of-the-art UNEt TRansformers (UNETR) of the Medical Open Network for Artificial Intelligence (MONAI) framework. We trained and tested our models using 618 de-identified Cone-Beam Computed Tomography (CBCT) volumetric images of the head acquired with several parameters from different centers for a generalized clinical application. Our results on a 5-fold cross-validation showed high accuracy and robustness with a Dice score up to 0.962±0.02. Our code is available on our public GitHub repository.

A Semi-Automatic Approach for Holistic 3D Assessment of Temporomandibular Joint Changes

“…Recently, fully automated image segmentation of the craniomaxillofacial bones from CBCT has been proposed [ 36 , 37 ]. These studies applied artificial intelligence, more specifically deep learning by neural networks, thereby eliminating the need of manual image segmentation and operator variability [ 36 , 37 ]. Artificial intelligence has been applied for automatic segmentation of the mandibular ramus and condyle [ 38 ].…”

Evaluating the Accuracy and Reliability of Blowout Fracture Area Measurement Methods: A Review and the Potential Role of Artificial Intelligence

“…Overall, automatic methods, such as computer-aided measurements and CT-based volumetric analysis, demonstrated higher accuracy and reliability compared with manual techniques. For example, Gillot et al 27 (2022) compared manual and automatic measurements of orbital fracture area using a 3-dimensional UNETR segmentation method on cone-beam CT scans and found that the automatic method showed higher accuracy and lower interobserver variability than the manual method. Inoue et al 28 (2022) developed an automated fracture screening method using an object detection algorithm on whole-body trauma CT, suggesting the potential advantages of automatic methods in detecting fractures with higher accuracy and reliability compared with manual techniques.…”

Automated artificial intelligence‐based three‐dimensional comparison of orthodontic treatment outcomes with and without piezocision surgery

“…In this study, we evaluated the tooth movement of all lower teeth after correcting mild to moderate malocclusion, while the literature mainly reports different evaluations and methodologies, such as the levelling of only mandibular anterior teeth, 27 assessment of severe maxillary malocclusion, 6 evaluation of maxillary canine distalization, 14 en-masse retraction utilizing miniscrews, 35 transversal tooth movement of mandibular lateral segments, 15 second molar protraction and upper canine retraction. 36 The automated AI-based dental tools used in this study are accurate 8,[19][20][21][22][23][24]26,31,32 and facilitate the assessment and quantification of tooth movement, reducing the time needed by clinicians and researchers to analyse imaging processes and evaluations by at least 90%. It is important to note that while commercial companies such as Relu, 37 Diagnocat 38 and Materialise, 39 as well as previous studies, have demonstrated similar applications, most of their tools are not integrated into the same platform and are not easily accessible due to cost and code unavailability.…”

“…The image pre-processing included T1 CBCT orientation, 15 T2 registration, 29 as well as T1 and T2 IOS registration to the CBCT scans 30 with validated semi-automated tools (Figure 1) and completely automated tools (Figure 2). 8,[19][20][21][22][23][24]26,31,32 The T1 CBCTs orientation was performed positioning the mandibular 3D model as follows: The lower border of the mandible, mesial surface of first molars and midline were aligned, respectively, with the axial, coronal and sagittal axis. 15 Subsequently, the T2 CBCT images were manually approximated to the T1 images and a voxel-based registration was performed.…”

“…The integration of these image modalities may optimize its diagnostic value, granting researchers and clinicians a comprehensive and enhanced analysis of the roots, root canals and crowns, 12,[17][18][19][20] also aiding in critical tooth position assessments. Additionally, the advent of innovative artificial intelligence (AI)-based tools [19][20][21][22][23][24][25][26] have facilitated merging root canals and crown segmentations, as well as automatic placement of landmarks and quantification of tooth movement.…”