Skip to main content

An artificial intelligence study: automatic description of anatomic landmarks on panoramic radiographs in the pediatric population



Panoramic radiographs, in which anatomic landmarks can be observed, are used to detect cases closely related to pediatric dentistry. The purpose of the study is to investigate the success and reliability of the detection of maxillary and mandibular anatomic structures observed on panoramic radiographs in children using artificial intelligence.


A total of 981 mixed images of pediatric patients for 9 different pediatric anatomic landmarks including maxillary sinus, orbita, mandibular canal, mental foramen, foramen mandible, incisura mandible, articular eminence, condylar and coronoid processes were labelled, the training was carried out using 2D convolutional neural networks (CNN) architectures, by giving 500 training epochs and Pytorch-implemented YOLO-v5 models were produced. The success rate of the AI model prediction was tested on a 10% test data set.


A total of 14,804 labels including maxillary sinus (1922), orbita (1944), mandibular canal (1879), mental foramen (884), foramen mandible (1885), incisura mandible (1922), articular eminence (1645), condylar (1733) and coronoid (990) processes were made. The most successful F1 Scores were obtained from orbita (1), incisura mandible (0.99), maxillary sinus (0.98), and mandibular canal (0.97). The best sensitivity values were obtained from orbita, maxillary sinus, mandibular canal, incisura mandible, and condylar process. The worst sensitivity values were obtained from mental foramen (0.92) and articular eminence (0.92).


The regular and standardized labelling, the relatively larger areas, and the success of the YOLO-v5 algorithm contributed to obtaining these successful results. Automatic segmentation of these structures will save time for physicians in clinical diagnosis and will increase the visibility of pathologies related to structures and the awareness of physicians.

Peer Review reports


Panoramic radiography is used for primary evaluations of the oral cavity for diagnosis of pathologies and treatment plans in dentistry [1]. It is possible to obtain information about the teeth, jawbones, sinuses, temporomandibular joints, and other hard tissues of the head and neck with panoramic radiography [2].

Panoramic radiographs, in which maxillary and mandibular anatomic structures can be observed, are used for the initial detection of cases closely related to the pediatric dentistry approach. Detection of dental anomalies and pathologies such as mesiodens, supernumerary teeth, odontoma, cysts, and impacted teeth on panoramic radiographs and determining their relationship with anatomic structures constitute a part of the first examination [3,4,5,6]. In pediatric patients for whom pain management is of great importance, injection of the local anesthetic solution near the mandibular foramen is a difficult anesthetic method to achieve, considering that the mandibular foramen position changes throughout the growth period [7]. Therefore, for a successful inferior alveolar nerve block, it is important to determine the position of the mandibular foramen correctly. In addition, the localization of the mandibular and mental foramen in children is an important factor to avoid damage to the neurovascular bundle in the management of complicated cases such as mandibular fractures in children [8, 9]. Mandible fractures constitute the majority (20–50%) of pediatric facial fractures, and condyle fractures are most common in the mandible. The risk of fracture is high, especially during the school-starting period. It is important to perform a panoramic X-ray in the first examination of these patients [10].

Artificial intelligence (AI) has been used for enhancing the analysis of dental radiology images. In the realm of two-dimensional (2D) radiographic images, digital radiographs are composed of myriad pixels, characterized by varying degrees of brightness and radiopacity. Artificial intelligence programs “learn” to analyze digital images by using these characteristics [11]. Deep learning is a machine learning method that excels in analyzing text and images [12]. Deep learning technique automatically learns the dataset using convolutional neural networks (CNN) which are a deep learning architecture, and creates a learning model after the learning process [13, 14]. There are many studies in the literature combining deep learning with dental radiography. Artificial intelligence technologies were used for the evaluation of pathological radiolucent lesions, diagnosis of osteoporosis, maxillary sinusitis, anatomic landmarks detection in adults and atherosclerotic plaques in carotid arteries on panoramic radiographs [15,16,17,18,19].

In some cases, when detailed examination of anatomic structures is required, it is possible to evaluate with three-dimensional imaging method, however, on clinical grounds, panoramic images are the primary available images for pediatric dentistry. Overlapping hard and soft anatomic landmarks (e.g. tongue, airway, nasal cavity flor, palate floor) on panoramic radiographs can make a difficult diagnosis [20]. Several factors contribute to the reliability of landmark identification in children: the definition of the landmark, the density and sharpness of images, the anatomic complexity of tissues, and the experience of the observers, especially for pediatric dentistry [2, 21].

Considering the advantages of showing the teeth and adjacent anatomic structures in a comprehensive way, being a fast and useful technique, easier application than intraoral techniques, and obtaining images with low radiation dose, AI-supported detection of pediatric anatomic landmarks in panoramic radiographs may prefer in the first examination in children. Artificial intelligence applications can serve as their guide to making better diagnoses and treatment plans with a low margin of error. It is thought that it will bring practicality to dentists [19]. To the best of our knowledge, although there have been a few studies on the mandibular canal and maxillary sinus, no comprehensive study has evaluated pediatric anatomic landmarks on panoramic radiographs using a deep-learning model. Therefore, the aim of the study was to investigate the success and reliability of the detection of anatomic landmarks observed on panoramic radiographs in children using artificial intelligence.


Data sources

In this study, in the pediatric patient population in the CranioCatch (Eskisehir, Turkey) database; A total of 981 mixed images for 9 different pediatric anatomic landmarks including maxillary sinus, orbita, mandibular canal, mental foramen, foramen mandible, incisura mandible, articular eminence, condylar and coronoid processes were labelled and were produced Pytorch-implemented YOLO-v5 models. This study was conducted in accordance with the code of ethics of the World Medical Association (Declaration of Helsinki). The approval of this study was granted by Eskisehir Osmangazi University Non-interventional Clinical Research Ethics Board, Eskişehir, Turkey which also approved informed consent (decision no. 04.10.2022/22). An informed consent was obtained from all subjects and/or their legal guardian(s).

Images of individuals in the pediatric population were included. Radiographic images obtained by incorrect positioning of the patient or containing motion artifacts were excluded from the study. Age, gender, and ethnicity differences were not observed, and the data was anonymized prior to uploading into the labeling system.

Labelling and training of data

Labelling is the process of identifying areas in an image and determining which region the object belongs to. Labelling; was carried out using web-based CranioCatch (Eskişehir, Turkey) software by oral and maxillofacial radiologist and a pediatric dentist with at least 10 years of experience. For labelling, anatomic regions were selected in the panoramic radiograph. The outer boundaries of these anatomical regions were established through polygonal segmentation and stored in JSON (JavaScript Object Notation) format, following a precise delineation process. Segmentation model 981 anonymized, mixed-sized panoramic radiography images were resized to 640 × 320 diameter. In this study, the training was executed employing the PyTorch library in Python. To ensure the integrity of the evaluation process, the dataset was partitioned into three distinct subsets: 80% designated for training purposes, 10% for validation to fine-tune parameters, and another 10% for final model testing.

  • Training group: 80% of the images (data set for training the model).

  • Validation group: 10% of the images (Denotes the set of instances that remain dissociated from the model’s training process and must remain concealed from the model’s purview throughout this phase. This subset is employed to assess the model, determining whether to halt the training or modify training parameters.)

  • Test group: 10% of the images (This subgroup was used to assess how well the trained model worked, using both the training and validation data.)

The training and validation datasets were important for figuring out the best settings for the AI algorithm. We then checked the success of the model performed using the test dataset.

Deep-learning algorithm

The images, after undergoing classification and labeling, were resized to dimensions of 640 × 320 pixels as part of the training process. In this study, the training phase was conducted using the PyTorch library within the Python programming language, employing 2D CNN architectures, and encompassing a span of 500 training epochs. For the segmentation task, the YOLO-v5 (Ultralytics, San Francisco, California, USA) was used. YOLO-v5, a member of the computer vision model lineage, offers a quartet of progressive versions, namely: small (s), medium (m), large (l), and extra-large (x). We used the YOLO-v5x model here. The architecture of YOLO-v5x comprises three fundamental constituents: Backbone, Neck, and Head. Notably, CSPDarknet53 serves as the foundational backbone. The Neck section integrates the Path Aggregation Network (PANet) and Spatial Pyramid Pooling (SPP) techniques, facilitating the fusion of extracted features. Subsequently, the YOLO Layer culminates in segmentation outcomes, encompassing class categorizations, confidence scores, and spatial coordinates. Upon the development of YOLO models, the PyTorch model’s outputs were subjected to assessment on a reserved 10% subset of the dataset, yielding insightful accuracy metrics.

The model’s effectiveness underwent assessment via a confusion matrix, a graphical depiction of the juxtaposition between predicted and actual situations. The model’s efficacy was gauged through the computation of performance metrics, derived from the interplay of True Positive (TP: accurate diagnoses correctly detected and segmented), False Positive (FP: diagnoses mistakenly identified and imprecisely segmented), and False Negative (FN: diagnoses incorrectly detected and segmented) evaluations. This evaluative framework encompassed metrics such as sensitivity, precision, and the F1 score—a harmonic medium of these indices.


14,804 labels a total of 981 mixed images were made for 9 different pediatric anatomic landmarks including maxillary sinus (1922), orbita (1944), mandibular canal (1879), mental foramen (884), foramen mandible (1885), incisura mandible (1922), articular eminence (1645), condylar (1733) and coronoid (990) processes. F1 score and sensitivity values for these anatomic regions were found as 0.98–0.99, 1–1, 0.97–0.99, 0.88–0.92, 0.95–0.95, 0.99–0.99, 0.92–0.92, 0.94–0.99, 0.86–0.97 respectively (Table 1).

Table 1 The sensitivity, precision, and F1 Score value of AI model estimation performance measures

True positive real and predictive images for pediatric anatomic landmarks are shown in Fig. 1. The estimation output obtained as a result of the detection of all landmarks by the AI model on a panoramic image is shown in Fig. 2.

Fig. 1
figure 1

True positive real (R) and predictive (P) images for pediatric anatomic landmarks (mandibular canal 1a-1b, maxillary sinus 2a-2b, mental foramen 3a-3b, condylar processes 4a-4b, foramen mandible 5a-5b, orbita 6a-6b, articular eminence 7a-7b, incisura mandible 8a-8b, coronoid process 9a-9b)

Fig. 2
figure 2

The estimation output of the AI model for all landmarks on a panoramic image


Panoramic radiography is the most commonly chosen imaging technique in which collaboration can be easily established in the pediatric patient group due to its simple and painless procedure. In this study, it aimed to evaluate the use of artificial intelligence technology to facilitate the detection of anatomic structures on panoramic radiographs in pediatric patients.

A systematic review reported that AI technology is widely used in different specialities of dentistry. AI technology has been used in oral and maxillofacial radiology, orthodontics, dentofacial orthopaedics, endodontics, periodontics, and oral and maxillofacial surgery [19]. CNNs have often been applied in orthodontics to help detect landmarks on lateral cephalometric radiographs [18, 22,23,24]. Dental diagnostic image analysis can also assist other medical professionals in diagnosing lesions and diseases [25,26,27,28]. To perform tooth numbering using deep learning algorithms on digital dental photographs, and to evaluate the success of these algorithms in determining the presence of frenulum, gingival hyperplasia and gingival inflammation. Yolov5 architecture were used in the creation of the models [29]. In the field of dentistry, the successful outcomes achieved in research utilizing artificial intelligence underscore its significant contributions, paving the way for transformative advancements in dental practice. These studies represent a crucial step towards accelerating the transformation of dentistry through the application of cutting-edge technology.

In addition to these advancements in dentistry, it’s worth noting that artificial intelligence (AI) technology has also become a new trend in forensic medicine, marking a watershed moment for the entire field of forensic science. In research aimed at facilitating interdisciplinary collaboration between forensic experts and deep learning engineers, a new workflow has been introduced for the 3D CNN analysis of full-head and neck cone-beam computed tomography (CBCT) scans. The application methods of 3D CNN for forensic investigations have yielded successful results in areas such as gender determination, estimation of biological age, annotation of 3D cephalometric landmark points, prediction of growth vectors, and estimation of facial soft tissue from the skull [30]. Age and gender assessment is used to identify victims, determine criminal responsibility, or identify individuals without legal documents [31, 32]. Studies using dental orthopantomogram examples, compared logistic regression linear models with neural networks for each legal age limit (14, 16, and 18 years) and showed that neural networks performed better [33, 34]. Similar studies in this field have favored neural networks for age prediction using dental orthopantomograms or landmarks in the mandible [35, 36]. It is certain that artificial intelligence will shape new developments and provide contributions in the field of forensic science.

Today, CBCT is a modern dental imaging system that provides rapid volumetric imaging with low radiation exposure. It has added a third dimension to the diagnostic process, surpassing manual and digital 2D analyses [37, 38]. Automated CBCT segmentation stands out as one of the current trends and innovations [39]. Undoubtedly, compared to the past when there were only 2D images such as panoramic or lateral X-rays, we have reached a higher level with CBCT and artificial intelligence. The benefits of using artificial intelligence for medical diagnosis in evaluating CBCT scans include improved accuracy, speed, and efficiency in diagnosis [40]. Recent studies evaluating the accuracy of AI-supported automatic detection of small edentulous areas in CBCT scans have shown that AI algorithms are fast and highly accurate in detecting teeth and small edentulous areas, tooth labelling is much faster, and segmentation is performed up to 900 times faster than when done by an experienced dental surgeon [41, 42].

There are some limitations to the processing of CBCT data by artificial intelligence [43], which can be divided into two main groups. Firstly, when considering the quality assessment of the literature, insufficient sample sizes and incomplete reporting are the main contributors to the high risk of bias. Secondly, AI techniques encompass highly heterogeneous applications in terms of dataset acquisition, analysis, and performance metrics, making comparisons challenging. Additionally, limitations such as AI algorithms affected by metallic artifacts [44, 45], variations in canal calcifications, errors due to examiner experience [46], and minimal cross-sectional areas [42, 45] are present in the processing of CBCT data by artificial intelligence.

Within the realm of dentistry, an investigation delved into the utilization of deep learning to segment the mandibular canal in panoramic radiography, yielding notably precise findings (with an accuracy of 0.847). This remarkable precision underscores the advantages of training on two-dimensional images, given that the canal prominently features in the 2D panoramic image, and the resolution of panoramas exceeds that of CBCT [47]. In a study aimed at improving the visualization of the Inferior Alveolar Canal (IAC), the edges of dental panoramic images were enhanced using a novel structural filter. Candidate regions were selected from the enhanced image by the proposed Multi Hidden Layer Extreme Learning Machine Artificial Neural Network (MELMANN) model. Experimental results indicate that this method effectively delineated the IAC [48].

In another study evaluating the success of different deep learning methods in detecting the mandibular canal with CBCT in adult patients, three-dimensional U-net was found to be the most successful. CBCT is an imaging modality that offers three-dimensional depth away from superpositions, therefore, many anatomical and pathological features can be defined more easily, and it can provide less complex sectional images for deep learning methods [49]. However, it is not an imaging method that can be used frequently in terms of radiation dose in the pediatric population. In the present study, models of the mandibular canal in panoramic radiographs were developed with a new-generation deep learning method, and success metrics were found to be high.

In dental panoramic X-ray images, it’s harder to see the mental foramen for patients under 20 years old, when compared to patients who are over 20 years old. Researchers have achieved a 95.4% mean dice similarity coefficient (DSC) for patients between 7 and 20 years of age working on 1000 dental pantomograph with the U-net model [50]. A study for segmentation of the mental foramen from 112 digital Panoramic x-ray images used a single full CNN based on U-net. Researchers reported the highest precision value 71.13% in this study [51]. In our data, it was difficult to distinguish the mental foramen, so we performed the training with the labels we made in about half of the total data set. The precision of the mental foramen model was lower than the other models, but we got a score of 0.84% here as well. This result is also promising for cases with difficult model detection. We think that both the number of mental foramen labels in panoramic radiography and the YOLO-v5, which we used in the development of the models, increased the success rate.

In a study, the researhers discovered that a deep learning system’s ability to diagnose maxillary sinusitis using panoramic X-rays was comparable to radiologists and even better than dental assistants [16]. Another research used deep learning to find and categorize issues in the maxillary sinus through panoramic X-rays. Like our study, they achieved a 100% success rate in detecting both normal and inflamed maxillary sinuses, mirroring our sensitivity results. The results showed that the maxillary sinus can be reliably detected in clinical practice with this AI model [14].

Particularly in pediatric patients, condylar fractures and displacement of the fractured condyle are frequently encountered as an indirect effect of a blow to the symphysis due to a face-down fall. This position change may be overlooked in trauma patients, and problems such as being late for treatment and stabilization cause displaced fragments to fuse in abnormal positions. In a study of panoramic radiographs of 100 condyles with and without mandibular condyle fractures from two hospitals to evaluate deep learning models, the system has been recommended for diagnosing condylar fractures as high classification performance was obtained, with operating characteristic curve values of > 0.85 and sensitivity 83–85% of data from both hospitals [52]. YOLOv5 were trained to automate the placement of the bounding boxes to detect fractures in the radiographic images. CNN-based models detected mandibular fractures with a performance surpassing that of experts [53]. The present study including 745 panoramic images, 0.99 sensitivity and 0.94 F1 score indicated that the YOLO-v5 model can be used successfully for mandibular automatic condyle segmentation.

As a limitation of the study, success may be low due to reasons such as the superposition of tooth germs in the mental foramen region, mixed dentition, and the presence of different densities in the relevant region. By increasing and diversifying the number of data, the success in detecting mental foramen should be increased with further studies.


In this study, we developed YOLO-v5 models that will provide automatic detection of 9 important anatomic formations in approximately one thousand panoramic radiographs of pediatric patients. The success of correct detection of the models was found to be quite high when compared to the literature. Superposition of anatomic formations on panoramic radiographs may cause difficulties in dentomaxillofacial diagnosis. Knowing the anatomy helps to distinguish these structures from pathologies in the first place, then it is important in many ways such as clarifying the relationship of the related structures with the dental structures, creating awareness about the pathological conditions of these structures, monitoring the findings of the dental structures in these regions or the effects of the structures on the dental structures. The development of systems that automatically segment these structures in the graphs will play a role as clinical decision-support mechanisms, increase the awareness of physicians and save time at the same time.

Data Availability

The datasets generated and/or analysed during the current study are not publicly available due [The data were obtained from anonymized radiographs by labelling them in a way that they cannot be accessed by reverse engineering. An informed consent form was obtained for the use of data for scientific purposes. We do not have patient consent to share raw data, in addition, the data is anonymized to not contain personal information.] but are available from the corresponding author on reasonable request.



Convolutional neural networks


Artificial intelligence




JavaScript Object Notation


Path Aggregation Network


Spatial Pyramid Pooling


True positive


False positive


False negative


Cone beam computed tomography


Dice similarity coefficient


Inferior Alveolar Canal


  1. Yüksel AE, Gültekin S, Simsar E, Gündoğar S, Tokgöz M, Hamamci S, et al. Dental enumeration and multiple treatment detection on panoramic X-rays using deep learning. Sci Rep. 2021;11:12342.

    PubMed  PubMed Central  Google Scholar 

  2. Cantekin K, Sekerci AE, Miloglu O, Buyuk SK. Identification of the mandibular landmarks in a pediatric population. Med Oral Patol Oral Cir Bucal. 2014;19(2):136–41.

    Google Scholar 

  3. Bekiroglu N, Mete S, Ozbay G, Yalcinkaya S, Kargul B. Evaluation of panoramic radiographs taken from 1,056 Turkish children. Niger J Clin Pract. 2015;18(1):8–12.

    PubMed  Google Scholar 

  4. Kuru S, Açıkgöz MM, Erdem AP, Ak G. Evaluation of maxillary sinus expansion in children due to maxillary first molar extraction. Eur Oral Res. 2019;53(1):1–5.

    PubMed  PubMed Central  Google Scholar 

  5. Margot R, Maria CDLP, Ali A, Annouschka L, Anna V, Guy W. Prediction of maxillary canine impaction based on panoramic radiographs. Clin Exp Dent Res. 2020;6(1):44–50.

    PubMed  Google Scholar 

  6. Ahn Y, Hwang J, Jung YH, Jeong T, Shin J. Automated mesiodens classification system using deep learning on panoramic radiographs of children. Diagnostics. 2021;11(8):1477.

    PubMed  PubMed Central  Google Scholar 

  7. Krishnamurthy NH, Unnikrishnan S, Ramachandra JA, Arali V. Evaluation of relative position of mandibular foramen in children as a reference for inferior alveolar nerve block using orthopantamograph. J Clin Diagn Res. 2017;11(3):ZC71–4.

    PubMed  PubMed Central  Google Scholar 

  8. Movahhed T, Makarem A, Imanimoghaddam M, Anbiaee N, Sarrafshirazi AR, Shakeri MT. Locating the Mandibular Foramen relative to the Occlusal plane using panoramic radiography. J Appl Sci. 2011;11:573–8.

    Google Scholar 

  9. Apaydın BK. Çocuklarda mandibular foramenlerin panoramik radyograflardaki konumu ve okluzal düzlemle ilişkisi. Selcuk Dent J. 2020;7(1):54–8.

    Google Scholar 

  10. Ghasemzadeh A, Mundinger GS, Swanson EW, Utria AF, Dorafshar AH. Treatment of Pediatric Condylar fractures: a 20-Year experience. Plast Reconstr Surg. 2015;136(6):1279–88.

    PubMed  PubMed Central  Google Scholar 

  11. Chen YW, Stanley K, Att W. Artificial intelligence in dentistry: current applications and future perspectives. Quintessence Int. 2020;51(3):248–57.

    PubMed  Google Scholar 

  12. Lecun Y, Bengio Y, Hinton G. Deep learning. Nature. 2015;521(7553):436–44.

    PubMed  Google Scholar 

  13. Heo MS, Kim JE, Hwang JJ, Han SS, Kim JS, Yi WJ, et al. Artificial intelligence in oral and maxillofacial radiology: what is currently possible? Dentomaxillofac Radiol. 2020;50(3):20200375.

    PubMed  PubMed Central  Google Scholar 

  14. Kuwana R, Ariji Y, Fukuda M, Kise Y, Nozawa M, Kuwada C, et al. Performance of deep learning object detection technology in the detection and diagnosis of maxillary sinus lesions on panoramic radiographs. Dentomaxillofac Radiol. 2021;50(1):20200171.

    PubMed  Google Scholar 

  15. Lee JS, Adhikari S, Liu L, Jeong HG, Kim H, Yoon SJ. Osteoporosis detection in panoramic radiographs using a deep convolutional neural network-based computer-assisted diagnosis system: a preliminary study. Dentomaxillofac Radiol. 2019;48(1):20170344.

    PubMed  Google Scholar 

  16. Murata M, Ariji Y, Ohashi Y, Kawai T, Fukuda M, Funakoshi T, et al. Deep-learning classification using convolutional neural network for evaluation of maxillary sinusitis on panoramic radiography. Oral Radiol. 2019;35(3):301–7.

    PubMed  Google Scholar 

  17. Lee A, Kim MS, Han SS, Park PG, Lee C, Yun JP. Deep learning neural networks to differentiate Stafne’s bone cavity from pathological radiolucent lesions of the mandible in heterogeneous panoramic radiography. PLoS ONE. 2021;16(7 July).

  18. Corbella S, Srinivas S, Cabitza F. Applications of deep learning in dentistry. Oral Surg Oral Med Oral Pathol Oral Radiol. 2021;132(2):225–38.

    PubMed  Google Scholar 

  19. Khanagar SB, Al-ehaideb A, Maganur PC, Vishwanathaiah S, Patil S, Baeshen HA, et al. Developments, application, and performance of artificial intelligence in dentistry – a systematic review. J Dent Sci. 2021;16(1):508–22.

    PubMed  Google Scholar 

  20. Perschbacher S. Interpretation of panoramic radiographs. Aust Dent J. 2012;57:40–5.

    PubMed  Google Scholar 

  21. Houston WJB, Maher RE, McElroy D, Sherriff M. Sources of error in measurements from cephalometric radiographs. Eur J Orthod. 1986;8(3):149–51.

    PubMed  Google Scholar 

  22. Park JH, Hwang HW, Moon JH, Yu Y, Kim H, Her SB, et al. Automated identification of cephalometric landmarks: part 1—Comparisons between the latest deep-learning methods YOLOV3 and SSD. Angle Orthod. 2019;89(6):903–9.

    PubMed  PubMed Central  Google Scholar 

  23. Niño-Sandoval TC, Guevara Pérez Sv, González FA, Jaque RA, Infante-Contreras C. Use of automated learning techniques for predicting mandibular morphology in skeletal class I, II and III. Forensic Sci Int 2017;281:187.e1-187.e7.

  24. Hwang HW, Park JH, Moon JH, Yu Y, Kim H, Her SB, et al. Automated identification of cephalometric landmarks: part 2-Might it be better than human? Angle Orthod. 2020;90(1):69–76.

    PubMed  Google Scholar 

  25. Miki Y, Muramatsu C, Hayashi T, Zhou X, Hara T, Katsumata A, et al. Classification of teeth in cone-beam CT using deep convolutional neural network. Comput Biol Med. 2017;80:24–9.

    PubMed  Google Scholar 

  26. Chen H, Zhang K, Lyu P, Li H, Zhang L, Wu J, et al. A deep learning approach to automatic teeth detection and numbering based on object detection in dental periapical films. Sci Rep. 2019;9(1):3840.

    PubMed  PubMed Central  Google Scholar 

  27. Tuzoff Dv, Tuzova LN, Bornstein MM, Krasnov AS, Kharchenko MA, Nikolenko SI et al. Tooth detection and numbering in panoramic radiographs using convolutional neural networks. Dentomaxillofac Radiol. 2019;48(4).

  28. Xu X, Liu C, Zheng Y. 3D tooth segmentation and labeling using deep convolutional neural networks. IEEE Trans Vis Comput Graph. 2019;25(7):2336–48.

    PubMed  Google Scholar 

  29. Kurt Bayrakdar S, Uğurlu M, Yavuz MB, Sali N, Bayrakdar İŞ, Çelik Ö, et al. Detection of tooth numbering, frenulum attachment, gingival overgrowth, and gingival inflammation signs on dental photographs using convolutional neural network algorithms: a retrospective study. Quintessence Int. 2023;54(8):680–93.

    PubMed  Google Scholar 

  30. Thurzo A, Kosnáčová HS, Kurilová V, Kosmeľ S, Beňuš R, Moravanský N, et al. Use of Advanced Artificial Intelligence in Forensic Medicine, Forensic Anthropology and clinical anatomy. Healthcare. 2021;9(11):1545.

    PubMed  PubMed Central  Google Scholar 

  31. Armanious K, Abdulatif S, Bhaktharaguttu AR, Küstner T, Hepp T, Gatidis S et al. Organ-Based Chronological Age Estimation Based on 3D MRI Scans. In Proceedings of the 28th European Signal Processing Conference, Amsterdam, The Netherlands, 24–28 August 2021; pp. 1225–1228.

  32. Sykes L, Bhayat A, Bernitz H. The effects of the Refugee Crisis on Age Estimation Analysis over the past 10 years: a 16-Country survey. Int J Environ Res Public Health. 2017;14(6):630.

    PubMed  PubMed Central  Google Scholar 

  33. Gu YC, Han M, Chi Y, Long H, Zhang D, Yang J, et al. Accurate age classification using Manual Method and deep convolutional neural network based on Orthopantomogram images. Int J Leg Med. 2021;135(4):1589–97.

    Google Scholar 

  34. Farhadian M, Salemi F, Saati S, Nafisi N. Dental Age Estimation using the pulp-to-tooth ratio in canines by neural networks. Imaging Sci Dent. 2019;49(1):19–26.

    PubMed  PubMed Central  Google Scholar 

  35. Štepanovský M, Ibrová A, Buk Z, Velemínská J. Novel age estimation model based on development of Permanent Teeth compared with Classical Approach and other Modern Data Mining methods. Forensic Sci Int. 2017;279:72–82.

    PubMed  Google Scholar 

  36. Vila-Blanco N, Varas-Quintana P, Aneiros-Ardao Á, Tomás I, Carreira MJ. Automated description of the Mandible shape by Deep Learning. Int J Comput Assist Radiol Surg. 2021;16(12):2215–24.

    PubMed  PubMed Central  Google Scholar 

  37. Lascala CA, Panella J, Marques MM. Analysis of the Accuracy of Linear measurements obtained by Cone Beam Computed Tomography (CBCT-NewTom). Dentomaxillofacial Radiol. 2014;33(5):291–4.

    Google Scholar 

  38. Thurzo A, Javorka V, Stanko P, Lysy J, Suchancova B, Lehotska V, et al. Digit Man Cephalometric Anal Bratisl Med J. 2010;111(2):97–100.

    Google Scholar 

  39. Urban R, Haluzová S, Strunga M, Surovková J, Lifková M, Tomášik J, et al. AI-Assisted CBCT Data Management in Modern Dental Practice: benefits, limitations and innovations. Electronics. 2023;12(7):1710.

    Google Scholar 

  40. Innovative Artificial Intelligence Applications in Dentistry. Available online: (accessed on 7 October 2023).

  41. Gerhardt MDN, Fontenele RC, Willems H, Jacobs R. Accuracy of an Artificial Intelligence-Driven Tool for the detection of small edentulous regions on Cone-Beam Computed Tomography. J Dent. 2022;121:103989.

    Google Scholar 

  42. Chung EJ, Yang BE, Byun SH, Yi S, Kim YH, Kang SH. Effectiveness of Cone-Beam Computed Tomography (CBCT)-Generated cephalograms using Artificial Intelligence (AI) Cephalometric Analysis. Sci Rep. 2022;12(1):20585.

    PubMed  PubMed Central  Google Scholar 

  43. Muresanu S, Almasan O, Hedesiu M, Diosan L, Dinu C, Jacobs R. Artificial Intelligence models for clinical usage in Dentistry with a focus on Dentomaxillofacial CBCT: a systematic review. Oral Radiol. 2023;39(1):18–40.

    PubMed  Google Scholar 

  44. Tsolakis IA, Kolokitha OE, Papadopoulou E, Tsolakis AI, Kilipiris EG, Palomo JM. Artificial Intelligence as an aid in CBCT Airway Analysis: a systematic review. Life. 2022;12(11):1894.

    PubMed  PubMed Central  Google Scholar 

  45. Albitar L, Zhao T, Huang C, Mahdian M. Artificial Intelligence (AI) for detection and localization of Unobturated Second Mesial Buccal (MB2) canals in Cone-Beam Computed Tomography (CBCT). Diagnostics. 2022;12(12):3214.

    PubMed  PubMed Central  Google Scholar 

  46. How Accurate Are Facial Recognition Systems—And Why Does It Matter?| Strategic Technologies Blog|CSIS. Available online: (accessed on 7 October 2023).

  47. Vinayahalingam S, Xi T, Bergé S, Maal T, de Jong G. Automated detection of third molars and mandibular nerve by deep learning. Sci Rep. 2019;9(1):9007.

    PubMed  PubMed Central  Google Scholar 

  48. Maheswari PU, Banumathi A, Ulaganathan G, Yoganandha R. Inferior alveolar nerve canal segmentation by local features based neural network model. IET Image Process. 2022;16(3):703–16.

    Google Scholar 

  49. Kwak GH, Kwak EJ, Song JM, Park HR, Jung YH, Cho BH, et al. Automatic mandibular canal detection using a deep convolutional neural network. Sci Rep. 2020;10(1):5711.

    PubMed  PubMed Central  Google Scholar 

  50. Dasanayaka C, Dharmasena B, Bandara WR, Dissanayake MB, Jayasinghe R. Segmentation of Mental Foramen in Dental Panoramic Tomography using Deep Learning. 2019:81 – 4.

  51. Kats L, Vered M, Blumer S, Kats E. Neural Network Detection and Segmentation of Mental Foramen in panoramic imaging. J Clin Pediatr Dent. 2020;44(3):168–73.

    PubMed  Google Scholar 

  52. Nishiyama M, Ishibashi K, Ariji Y, Fukuda M, Nishiyama W, Umemura M, et al. Performance of deep learning models constructed using panoramic radiographs from two hospitals to diagnose fractures of the mandibular condyle. Dentomaxillofac Radiol. 2021;50:20200611.

    PubMed  PubMed Central  Google Scholar 

  53. Warin K, Limprasert W, Suebnukarn S, Inglam S, Jantana P, Vicharueang S. Assessment of deep convolutional neural network models for mandibular fracture detection in panoramic radiographs. Int J Oral Maxillofac Surg. 2022;51(11):1488–94.

    PubMed  Google Scholar 

Download references


Not applicable.


There is no funding to report for this manuscript.

Author information

Authors and Affiliations



I.B., E.B., I.S.B., and K.O. design the work; I.B. and E.B. interpretation of data and write the manuscript; E.B., I.S.B., O.B., F.M.A., and O.C. developed the artificial intelligence model, I.B., E.B., and K.O. drafted the work or substantively revised it. All authors read and approved the final manuscript.

Corresponding author

Correspondence to İrem Bağ.

Ethics declarations

Ethics approval and consent to participate

This study was conducted in accordance with the code of ethics of the World Medical Association (Declaration of Helsinki). The approval of this study was granted by Eskisehir Osmangazi University Non-interventional Clinical Research Ethics Board, Eskişehir, Turkey which also approved informed consent (decision no. 04.10.2022/22). An informed consent was obtained from all subjects and/or their legal guardian(s).

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.

Additional information

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit The Creative Commons Public Domain Dedication waiver ( applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Bağ, İ., Bilgir, E., Bayrakdar, İ.Ş. et al. An artificial intelligence study: automatic description of anatomic landmarks on panoramic radiographs in the pediatric population. BMC Oral Health 23, 764 (2023).

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: