Chairs:
kristina hellén-halme
ruben pauwels
263: 3D JAW BONE SEGMENTATION WITH ARTIFICIAL INTELLIGENCE
U.E. Karaturgut1, E. Bilgir2, Ö. Çelik3, İ.Ş. Bayrakdar2
1Beykent University/Faculty of Dentistry, Department of Dentomaxillofacial Radiology, İstanbul, Turkey, 2Eskişehir Osmangazi University / Faculty of Dentistry, Department of Dentomaxillofacial Radiology, Eskişehir, Turkey, 3Eskişehir Osmangazi University, Faculty of Science, Department of MathematicsComputer, Eskişehir, Turkey
Aim: In implant planning and guide production, the distance of the crests to the anatomical areas is taken as basis. Deep learning is a subset of artificial intelligence where algorithms are designed to mimic the structure and function of the human brain to process data and make decisions. In dentistry, deep learning plays a crucial role in image analysis, treatment planning, and patient care. The aim of this study is to evaluate the success of artificial intelligence models in three-dimensional automatic segmentation of the maxillary alveolar bone and mandible.
Matherial and Methods: In the study, 34551 labels were made on 165 dicom images. 90% of the data is allocated as training and 10% as testing. Trainings were carried out with nn-Unet. The success of the models was determined by the dice score.
Results: The dice score was found to be 0.98 for the mandible and 0.89 for the maxilla alveolar process.
Conclusion: Automatic segmentation of the jaw bones is the first step of guide preparation. The findings we obtained are promising for artificial intelligence-supported guide applications.
262: IMAGING DIAGNOSIS OF TUMOR MASSES IN THE NASAL CAVITIES IN CHILDREN AND YOUNG ADULTS
D. Haba1, C. Antohi1, A. Nemtoi2, R. Popescu1, E. Marciuc1, D.-A. Ilinca1, M.S.C. Haba3, A.E. Sirghe1
1University Grigore T. Popa Iasi, Surgery, Iasi, Romania, 2University „ Stefan cel Mare“, Department of Biomedical Sciences, Suceava, Romania, 3University Grigore T. Popa Iasi, Medicine II, Iasi, Romania
Aim: Tumor masses in the nasal cavity are quite common. They can have various etiologies: congenital, inflammatory, infectious, traumatic or malignant. Very few of these etiologies are unique to the pediatric population such as encephalocele, nasal glioma, juvenile nasopharyngeal angiofibroma. s.a. The purpose of this study is to present the imaging aspects encountered in nasal cavity lesions in children and young adults in the NE region of Romania.
Matrial and Methods: In the last two years (Jan. 2022- Jan. 2024), a total of 123 paediatric patients were addressed to Medimagis Clinic for further imagistic investigations. We analised 50 patients via CBCT, 60 patients with computer-tomography, and we complete this study with 13 paediatric patients in MRI Emergency N. Oblu Hospital registers for a period of 5 years.
Results: Non-neoplastic disorders were identified in 62 femele patients and 48 male patients mostly with infectious, inflamatory and traumatic lesions. Only 6 male patients had juvenile nasopharyngeal angiofibroma, two nasal pleomorphic adenoma, one infantile hemangioma and only 7 cases had maglignat neoplasm- two rabdomyosrcoma, two lymphoma, one nasopharyngeal carcinoma, one metastatic disease and one esthesioneuroblastoma.
Conclusion: CBCT and CT scan are essential in evaluating congenital, inflammatory, traumatic or tumor lesions to reveal bone extension of the lesion. MRI examination is essential for characterizing the tumor mass and perineural extension of the neoplazic disorders.
267: ADVANCEMENTS IN ORBITAL DEFECT RECONSTRUCTION PLANNING: A STATISTICAL SHAPE MODEL APPROACH OUTPERFORMS TRADITIONAL MIRRORING METHODS
J. Van Dessel1, P. Vanslambrouck1,2, C. Politis1, R. Willaert1, M. Bila1, P. Claes3,4,5, Y. Sun1
1Oral and Maxillofacial Surgery, University Hospitals Leuven and OMFS-IMPATH Research Group, Department of Imaging & Pathology, Faculty of Medicine, KU Leuven, Leuven, Belgium, 2Department of Computer Science, KU Leuven, Leuven, Belgium, 3Department of Electrical Engineering, ESAT/PSI, KU Leuven, Leuven, Belgium, 4Department of Human Genetics, KU Leuven, Leuven, Belgium, 5Medical Imaging Research Center, University Hospitals Leuven, Leuven, Belgium
Aim: Traditional methods for reconstructing orbital bone defects rely on manual mirroring techniques, which are time-consuming and subject to variability based on the expertise of the surgical planner. This study aims to introduce a novel approach utilizing a statistical shape model (SSM) for automated and precise orbital defect reconstruction.
Material and Methods: A SSM was constructed in open-source Scalismo software, using computed tomography (CT) scans from sixty-five healthy midfaces to capture the shape variations of the orbital region. Parameter optimization and clinical validation were conducted on an additional fifteen CT scans. Orbital defects, ranging in size from large to small and occurring both unilaterally and bilaterally, were simulated within orbital floor and medial wall regions. The reconstruction error for these defects was then compared between the SSM-based reconstruction and the traditional mirroring method.
Results: The SSM-based reconstruction method achieved median reconstruction errors of 0.26mm for small and 0.35mm for large unilateral defects, as well as 0.32mm for small and 0.38mm for large bilateral defects. These results outperformed those of the mirroring method, which yielded a reconstruction error of 0.48mm for small and 0.51mm for large unilateral defects. Notably, the mirroring method was found to be inaccurate for bilateral defects. The SSM-based reconstruction had a mean processing time of 2±2mins. In constrast, experienced engineers spent an average of 25±11mins.
Conclusion: SSM-based reconstruction offers superior accuracy and efficiency compared to mirroring techniques, especially for bilateral defects. These findings highlight the potential of SSM-based reconstruction to revolutionize patient-specific implant design protocols.
249: ARTIFICIAL INTELLIGENCE IN PANORAMIC RADIOGRAPHY INTERPRETATION: A GLIMPSE INTO THE RADIOGRAPHIC EXAMINATION OF THE FUTURE
A. Kuran1, E. Bilgir2, Ö. Çelik2, İ.Ş. Bayrakdar2
1Kocaeli University, Kocaeli, Turkey, 2Eskişehir Osmangazi University, Eskişehir, Turkey
Aim: The objective of this study is to develop a deep learning algorithm based on YOLO-v8 for the purpose of diagnostic charting in panoramic radiography, followed by a comprehensive evaluation of its performance.
Material and Method: Panoramic radiographs were labelled using CranioCatch labelling software (Eskisehir, Turkey) in various categories including adult tooth numbering, caries, unerupted/impacted tooth, fillings, overhanging fillings, root canal filling, periapical lesions, residual root, alveolar bone loss, vertical bone loss, horizontal bone loss, furcation defect, dental implant, fixed prosthesis (pontic, crown, implant-supported crown), jaw pathology and dental pulp. The model was trained using 389056 labels and tested using 48346 labels. A YOLO-v8 model was used for developing Artificial Intelligence (AI) model for each labelling. Confusion matrix was used to success evaluation of AI models.
Results: The performance metrics for the developed model were as follows: precision/sensitivity/F1-score for adult tooth numbering (0.99/0.99/0.99), fillings (0.99/0.97/0.98), vertical bone loss (0.77/0.68/0.72), dental implants (0.99/0.99/0.99), horizontal bone loss (0.98/0.86/0.93), dental pulp (0.99/0.97/0.98), furcation defects (0.91/0.87/0.89), fixed prostheses (0.97/0.96/0.97), root canal fillings (0.99/0.98/0.99), alveolar bone loss (0.94/0.86/0.90), jaw pathologies (0.93/0.78/0.85), periapical lesions (0.96/0.66/0.78), residual roots (0.93/0.97/0.95), caries (0.95/0.91/0.93), unerupted/impacted tooth (0.85/0.96/0.90), and overhanging fillings (0.86/0.91/0.89).
Conclusion: Upon scrutiny of the outcomes derived from the developed models, it is discerned that convolutional neural network-based AI exhibits a notable capacity to identify various conditions encountered during routine clinical examinations in panoramic radiographs, demonstrating high performance. Aligned with these findings, it can be asserted that AI holds substantial promise for physicians in enhancing routine clinical practices.
254: YOLO-V5 BASED DEEP LEARNING APPROACH FOR TOOTH DETECTION AND SEGMENTATION ON PEDIATRIC PANORAMIC RADIOGRAPHS IN MIXED DENTITION
B. Beşer1, T. Reis2, M.N. Berber1, E. Topaloğlu3, E. Güngör3, M.Ç. Kılıç4, S. Duman3, Ö. Çelik5, A. Kuran6, İ.Ş. Bayrakdar5
1Recep Tayyip Erdogan University, Rize, Turkey, 2Private Practice, Trabzon, Turkey, 3Inonu University, Malatya, Turkey, 4Beykent University, Istanbul, Turkey, 5Eskişehir Osmangazi University, Eskişehir, Turkey, 6Kocaeli University, Kocaeli, Turkey
Aim: In the interpretation of panoramic radiographs, the identification and numbering of teeth is an important part of the correct diagnosis. This study evaluates the effectiveness of YOLO-v5 in the automatic detection, segmentation, and numbering of deciduous and permanent teeth in mixed dentition pediatric patients based on panoramic radiographs (PRs).
Methods: A total of 3854 mixed pediatric patients PRs were labelled for deciduous and permanent teeth using the CranioCatch labeling program. The dataset was divided into three subsets: training (n=3093, 80% of the total), validation (n=387, 10% of the total) and test (n=385, 10% of the total). An artificial intelligence (AI) algorithm using YOLO-v5 models were developed.
Results: The sensitivity, precision, F-1 score, and mean average precision-0.5 (mAP-0.5) values were 0.99, 0.99, 0.99, and 0.98 respectively, to teeth detection. The sensitivity, precision, F-1 score, and mAP-0.5 values were 0.98, 0.98, 0.98, and 0.98, respectively, to teeth segmentation.
Conclusions: YOLO-v5 based models can have the potential of to detect and segmentation of deciduous and permanent teeth accurately in PRs of pediatric patients with mixed dentition.