Advancing Dental Diagnostics: A Review of Artificial Intelligence Applications and Challenges in Dentistry

: The rise of artificial intelligence has created and facilitated numerous everyday tasks in a variety of industries, including dentistry. Dentists have utilized X-rays for diagnosing patients’ ailments for many years. However, the procedure is typically performed manually, which can be challenging and time-consuming for non-specialized specialists and carries a significant risk of error. As a result, researchers have turned to machine and deep learning modeling approaches to precisely identify dental disorders using X-ray pictures. This review is motivated by the need to address these challenges and to explore the potential of AI to enhance diagnostic accuracy, efficiency, and reliability in dental practice. Although artificial intelligence is frequently employed in dentistry, the approaches’ outcomes are still influenced by aspects such as dataset availability and quantity, chapter balance, and data interpretation capability. Consequently, it is critical to work with the research community to address these issues in order to identify the most effective approaches for use in ongoing investigations. This article, which is based on a literature review, provides a concise summary of the diagnosis process using X-ray imaging systems, offers a thorough understanding of the difficulties that dental researchers face, and presents an amalgamative evaluation of the performances and methodologies assessed using publicly available benchmarks.


Introduction
Artificial intelligence (AI) represents a significant technological advancement, enabling not only robots, but also various AI systems, to emulate intelligent human behavior.In therapeutic contexts, the concept of enhanced intelligence extends AI's application within the medical profession, aiming to improve accuracy and efficacy in dental diagnostics.Augmented intelligence emphasizes AI's supportive and complementary role alongside medical specialists, leveraging its extensive data-processing skills to assess results effectively [1].
X-ray exams aid in dental diagnosis by allowing oral professionals to access interior tooth structures that may not be apparent with an eye by lessening the visual diagnostic load.Various dental X-rays have completely changed dentistry by giving precise insights into the architecture of the mouth and teeth.With the use of these imaging technologies, dentists may detect gum disease, tooth decay, and other disorders early on, avoiding more problems and improving treatment results.Moreover, dental X-rays might reveal hidden problems that may be challenging to find on a clinical examination, such as impacted teeth [2].As artificial intelligence is now widely used in dentistry, it has become a valuable tool for both detecting and forecasting the course of illness.It is certain that artificial intelligence and information systems will accurately diagnose common ailments.Furthermore, while patients may trust a dental diagnosis, they may also harbor concerns about how it operates.Therefore, it remains crucial for systems to access sufficiently large datasets, maintain balanced separation, and effectively analyze data to uphold quality standards.Furthermore, ethical considerations must be addressed when utilizing data in AI technologies, aligning with the need for transparency in how these technologies function from the patient's perspective [3].
With an emphasis on the years beginning in 2010, this scoping study seeks to explore the state of AI-assisted diagnosis in oral health using X-ray images.The goals are to provide an overview of the area's present level of growth, highlight its limits, and pinpoint the research gaps that must be addressed in order to advance the area.
Among the principal contributions are: • Providing a schema for the current X-ray imaging literature; Reviewing the state-of-art of AI models used in dental practice; • Analyzing the possibilities, limitations, and future trends in using artificial intelligence in dentistry.
The remaining part of this paper is organized as follows: Section 2 provides an overview of the materials and some methods employed in the study; In Section 3, we delve into the challenges in automated diagnosis of dental diseases, outlining the problems encountered in this domain.;Section 4 explores the various types of dental diseases; Section 5 investigates the approaches to diagnosing dental diseases through X-ray imaging; Section 6 focuses on the crucial aspect of preprocessing to enhance the quality of data; Section 7, Section 8 explores the evolution of AI diagnostic tools in dentistry and addresses the datasets used in articles; Section 9 delves into the relevant Work Experiences; Section 10 delves into the specific role of X-ray imaging in dental diagnostics; in Section 11, we investigate the impact of AI on dental healthcare, exploring the transformative effects of artificial intelligence in this field; Section 12 delves into knowledge gaps and future research directions; and finally, Section 13 concludes with a summary and findings.

Protocol
The Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines were used in the study of the available literature in order to assess the reporting quality of systematic reviews.

Electronic Search Strategy
Google Scholar was used to conduct a thorough electronic search for every relevant research study in the database between 2010 and 2023.The search focused on publications related to the application of artificial intelligence, deep learning, and neural networks in dentistry.Specifically, we included the following types of materials:

•
Peer-reviewed journal articles; Timeline: Publications over the past 14 years (2010-2023) focused on the application of artificial intelligence, deep learning, and neural networks in dentistry; II.Language: All English-language publications were incorporated, regardless of the location of publication; III.Data and Outcome: Studies that adequately explain the datasets exploited, supported by explicit reporting of predictive and measurable outcomes to measure the effectiveness of the suggested model.

Exclusion Criteria
I.
Type of data used: research that does not provide precise details on the kinds of data that were used; II.Methodology: inadequately documented research on deep learning, machine learning, and computer vision techniques; III.Outcome: Studies that failed to document quantifiable results.

Study Selection and Items Collected
Following the elimination of duplicate papers, the title and abstract were reviewed.Eligibility criteria were used to analyze the papers' complete text.Finally, the references in the paper were carefully examined.

Challenges in Automated Dental Disease Diagnosis
AI-based models for dental disease detection, diagnosis, and prediction have become quite popular recently.However, challenges include aspects such as generality, accessibility, and restricted data availability.In order to address these challenges and assist dental professionals in treatment and prognosis planning, the primary focus of research continues to be on creating effective and precise solutions.AI models may struggle to achieve high sensitivity and specificity due to picture quality and anatomical structural differences [4].
• The significant quantity of data needed for training, validation, and testing is a sig- nificant obstacle in deploying artificial intelligence (AI) for dental caries detection.This involves extremely sensitive patient data, such as dental pictures and medical histories.Data sensitivity is a critical issue due to the personal and private nature of health information, which includes identifiable patient records and medical histories.The misuse of or unauthorized access to this data can lead to severe ethical and privacy violations.Ensuring data privacy and security is paramount to maintaining the trust of patients and upholding ethical standards in AI applications for dentistry diagnostics.Strict adherence to data protection regulations, such as GDPR, and implementing robust encryption and anonymization techniques are essential to safeguarding patient information [5].

•
One major concern revolves around the transparency of artificial intelligence algorithms and data.The reliability of AI systems' predictions greatly depends on the accuracy of annotations and labeling in the training dataset.Inaccurately labeled data can lead to subpar outcomes.This issue is especially pronounced in clinic-labeled datasets, where the lack of consistent quality further limits the transparency and effectiveness of AI systems [6].

Types of Dental Diseases
In Figure 1, we aim to provide a brief background of various oral health conditions discussed in the literature.These conditions encompass a spectrum of issues ranging from common concerns, like dental caries (tooth decay), to more intricate challenges, such as oral cancer.

Approaches to Diagnosing Dental Diseases through X-ray Imaging
This section delves into various research investigations aimed at identifying dental pathologies.Figure 2 depicts a process flow diagram that serves as a visual guide for the diagnostic procedures of dental ailments.

Approaches to Diagnosing Dental Diseases through X-ray Imaging
This section delves into various research investigations aimed at identifying dental pathologies.Figure 2

Approaches to Diagnosing Dental Diseases through X-ray Imaging
This section delves into various research investigations aimed at identifying den pathologies.Figure 2 depicts a process flow diagram that serves as a visual guide for t diagnostic procedures of dental ailments.

Preprocessing Techniques for Dental X-ray Pictures
Several preprocessing steps are performed to improve the quality of dental X-ray pictures and facilitate efficient interpretation before they are prepared for analysis.To diversify the dataset, methods such as data augmentation-which involves rotation, scaling, and other operations-are used.To comprehend visual aspects like texture, features such as entropy, intensity, and local binary patterns are extracted.
Resizing an image to a standard format and cropping it to remove information are examples of image modification steps.To remove the upper and lower jaws as well as the teeth and define a particular region of interest, segmentation is utilized.Binary pictures are produced by converting to RGB (red, green, and blue) format and then thresholding them.
Enhancement includes applying different filtering methods, quadtree decomposition, and contrast normalization.Pictures in the dataset that have missing teeth were culled.Refining the dataset involved flipping photos and segmenting them to reduce input sizes.The dataset was supplemented by machine learning techniques such as ImageDataGenerator, and certain photos were kept aside for validation.
The accuracy of dental X-ray detection is increased by techniques including picture sharpening and adaptive histogram equalization.Also, a comprehensive approach to preprocessing is taken, setting the stage for a later study that takes advantage of deep learning models like SqueezeNet and AlexNet.

Datasets Used for AI-Based Systems for Dentistry E-Health
This section includes details of the datasets used in some studies as a general review, starting from the earliest to the most recent paper.
Firstly, in 2017, Ref. [7] used experimental datasets from Hanoi Medical University in Vietnam.The dataset was divided into a training dataset and test dataset, and included 56 dental photographs taken between 2014 and 2015.
One year forward, in 2018, the Department of Periodontology at Daejeon Dental Hospital, Won Kwang University, conducted a study [8] that was authorized by the Institutional Review Board of the hospital.The dental hospital's PACS system provided the image datasets, which were gathered between January 2016 and December 2017.The images were categorized using electronic medical records.The training and validation datasets included 2400 photos (1200 dental caries and 1200 non-dental caries), whereas the test dataset contained 600 images (300 dental caries and 300 non-dental caries).The datasets were balanced in terms of dental caries against non-dental caries.Among the 3000 periapical radiographs in the dataset, approximately a quarter each represented maxillary premolars (25.9%) and maxillary molars (25.6%), while mandibular premolars accounted for 24.1%, and mandibular molars for 24.4%.Dental caries were detected in 29% of premolars and 27% of molars, whereas non-dental caries were observed in 26% of premolars and 23% of molars.
In the same year, Ref. [9] used 1500 panoramic images with considerable variability that were divided into 10 groups.The training data included 193 annotated images that were divided into two groups for training the segmentation network and validating the findings.After training and fine-tuning, the model was tested on 1224 dental arch pictures.Remarkably, the system utilized transfer learning from the Microsoft Common Objects in Context (MSCOCO) dataset to train itself using just 193 buccal pictures.
A year ahead, in 2019, in Ref. [10], a dataset of 1574 anonymized adult panoramic X-ray images was utilized.The images were randomly selected from the archival database at the Reutov Stemmatological Clinic in Russia, covering the period from January 2016 to March 2017.The panoramic X-ray images were captured using the Sirona OrthoPhos XG-3 X-ray apparatus, known for its technological advancements.To ensure methodological rigor, the dataset was divided into two distinct sections: (1) a primary training set and (2) a secondary testing set.The primary training set consisted of 1352 images and was used for the development and refinement of the algorithms for teeth detection and numbering.The secondary testing set comprised 222 images and was specifically employed to evaluate the accuracy and performance metrics of the developed software.This division underscores the meticulous approach taken, ensuring a comprehensive evaluation of the software's capabilities in automatic teeth detection and numbering.
Further, in the same year, the study [11] by Hu Chen et al. consisted of 1250 digitized dental periapical films as a dataset collected from Peking University School and Hospital of Stomatology.The films were anonymized and digitized at a resolution of 12.5 pixels per mm.An expert dentist annotated each tooth in the images with a bounding box and corresponding tooth number using the Federation Dentaire Internationale (FDI) numbering system.The collection of images was segmented into three parts for the study: a training set with 800 images, a validation set comprising 200 images, and a test set containing 250 images.
Furthermore, in 2019, the dataset in the study [12] by V. Geetha and K. S. Aprameya was obtained from SJM Dental, who compiled a dataset consisting of 49 dental X-ray images featuring caries and 16 images of healthy teeth.These images were captured using a Gendex X-ray unit equipped with an RVG sensor from Sirona.A professional dentist provided annotations of the carious lesions present in the dental radiographs.
Lastly, in 2019, the dataset in the study [13] by Shankeeth Vinayahalingam et al. was sourced from orthopantomograms (OPGs) of individuals who received treatment at the Oral and Maxillofacial Surgery Department of Radboud University Nijmegen Medical Centre in the year 2017.A set of 81 OPGs were randomly divided into two groups: 70% training and 30% validation/testing, and carefully chosen for analysis according to designated inclusion criteria; all images were de-identified to maintain patient confidentiality before the analysis commenced.
Shifting the focus to 2020, the dataset in the study [14] conducted by scholars from the Microelectronics CAD Center at Hangzhou Dianzi University, Hangzhou, China, and the West China School of Stomatology at Sichuan University, comprised 2491 dental Xrays for model training purposes and 138 X-rays for the validation process.The dental radiographs were sorted into six distinct groups, each corresponding to different sequences of tooth numbering.
In that year, the dataset in Ref. [15] by Wenzhe You et al. was gathered from the Department of Pediatric Dentistry at Peking University School and Hospital of Stomatology in Beijing, China.This collection included 886 intraoral photographs of primary teeth, with 709 photos allocated for training and 177 photos earmarked for testing.These images formed the basis for training an artificial intelligence model specifically tailored to the detection of dental plaque.
The dataset in a study [16] conducted in 2020 by Minyoung Chung et al. was sourced in Korea from Osstem Implant Co. Ltd.The collection of data included 818 panoramic dental X-ray images obtained from various dental clinics.Among these images, 574 were utilized for training, 162 were allocated for validation, and 82 were reserved for testing.
Further, in 2020, the dataset used in Ref. [17] consisted of approximately 2000 anonymized panoramic radiography images, collected from three dental clinics.From these, 1000 images were selected for detailed annotation.
In the year that followed, 2021, the dataset used in Ref. [18] was sourced from Kaggle.This dataset comprised visual images representing cavities and non-cavities.It contained a total of 74 images; 60 were used as training data and 14 were used as testing data.The training set covered 45 caries images and 15 non-caries images, while the testing set had 10 caries images and 4 non-caries images.This dataset was utilized for training and testing a deep learning model for classifying dental cavities.
Within the same year, the dataset used in Ref. [19] was specially collected for that research and sanctioned by the Institutional Review Board of National Yang-Ming University.It consisted of 100 teeth with various degrees of caries lesions.These teeth were then disinfected with 5% sodium hypochlorite and kept in distilled water.The dataset was narrowed down by excluding fractures, previously repaired teeth, and teeth with visible cavities.This procedure left 63 teeth for the purpose of examination.
Furthermore, focusing on this year, dental panoramic radiographs were taken at Asahi University Hospital as part of normal exams with the QRmaster-P system (Takara Telesystems Corp., Osaka, Japan).The images had a resolution of 3000×1500 pixels and a pixel size of 0.1mm.For that study, 100 pictures were used for training and assessment.Due to the limited dataset, four sets of photographs were randomly selected, and fourfold cross-validation was carried out.Three groups were used for training in each cycle, while the fourth group was used for assessment.For each of the four groups, this process was repeated [21].
Looking ahead to the upcoming year, a study conducted by Andac Imak et al. examined images derived from two private oral and dental health clinics; these images were captured using periapical radiography systems, namely Planmeca Intra from Helsinki, Finland, and MyRay from Imola, Italy.The analysis included 380 periapical pictures from 310 subjects.Forty images were excluded from the analysis because they were unclear or showed fractures, cysts, or infections.The total collection consisted of 340 periapical pictures.The patient-derived dataset included radiographs of 150 images depicting carious teeth, while 190 depicted non-carious teeth [22].
JK, an experienced dentist, used sophisticated equipment in 2022 to capture images for earlier research and documentation in the study "Caries Detection on Intraoral Images Using Artificial Intelligence".To avoid condensation, intraoral reflectors were utilized when photographing molar teeth.Quality control entailed eliminating insufficient or duplicated pictures.The images were modified for quality and standardized in size.The study concentrated on healthy or carious surfaces and excluded non-carious flaws.A total of 2417 high-quality clinical photographs were included, representing 1317 occlusal surfaces and 1100 smooth surfaces [23].
Furthermore, in 2022, panoramic X-rays (OPGs) obtained from clinics were incorporated into the study in [24].A DSLR was used for some of the OPGs, while clinics supplied digital copies for others.Each of the high-quality pictures, which made up a unique dataset of 1200 pictures, divided into 70% training and 30% testing images, showed people with various dental diseases and ages.
Exploring further within the year 2022, the dental image dataset used in Ref. [25] included panoramic dental X-ray images from a total of 116 people.Moreover, 90% of the data were taken as the training dataset, and 10% of the data were used for validation purposes.These images were obtained from the Medical Imaging Center in Iran and were anonymized and unidentified.The dataset encompassed various dental conditions, ranging from teeth in good condition to those with partial and complete edentulism.The different conditions of the cases were manually segmented by two dentists.
Reference [26], in 2022, involved 445 CBCT scans, from which 890 maxillary sinuses were examined.Manual segmentation was performed on the air space, MT, and MRCs in each sinus.High accuracy in detecting MT and MRCs was achieved by a three-step CNN algorithm trained on low-dose CBCT scans.The results demonstrated by the algorithm displayed comparable performance on both low-dose and full-dose CBCT scans.The findings suggest the CNN algorithm's potential for accurate detection and segmentation of MT and MRCs in the maxillary sinus using CBCT scans.
In 2022, Ref. [27] aimed to enhance the accuracy of diagnosing dental caries in children using panoramic radiographs.It utilized panoramic radiographs of 94 patients who did not have any caries and 210 patients who had one or more instances of dental caries.Of the total dataset of 6028 teeth, 3039 teeth had caries, while 2989 teeth were free from caries.The panoramic radiographs, formatted in JPEG, had an approximate size of 2441 × 1150 pixels.
Lastly, in 2022, the dataset used in Ref. [28] comprised a collective sum of 1432 images of dental conditions.These images were tagged and manually analyzed by an expert.
In a different context of the same year, Ref. [29] developed a framework using CNNs for diagnosing dental ailments in panoramic radiographs (PRs).The framework achieved high specificity for various dental diseases, such as caries, impacted teeth, missing teeth, residual roots, and full crowns.That study utilized a dataset of 2278 PRs, consisting of 1996 PRs for training and 282 PRs for evaluation, sourced from the Stomatology Hospital of Zhejiang Chinese Medical University (not publicly available due to privacy restrictions).It highlighted the potential of CNN-based AI in improving dental diagnostics using PRs, leveraging a large dataset for thorough assessment.
As for the papers not previously addressed, namely [30][31][32][33][34][35][36][37][38][39][40], it is evident that a significant limitation present in these studies is the lack of comprehensive details regarding the dataset employed.These articles notably omitted crucial information that would otherwise contribute to a more thorough understanding and evaluation of their respective research methodologies and findings.

Evolution of AI-Diagnostic Tools in Dentistry
A general overview of the AI techniques and architectures employed for various dental data types is provided as follows:

Convolutional Neural Networks (CNNs)
In dental applications, CNNs are extensively used to analyze image data.They are composed of convolution layers, pooling layers, and fully connected layers.The following are the primary CNN architectures used: An advanced object detection model that integrates a region proposal network (RPN) with a CNN.The RPN generates region proposals that are then classified and refined by CNN to detect objects and their bounding boxes within an image [7]. Figure 3 shows the Faster R-CNN structure.
high specificity for various dental diseases, such as caries, impacted teeth, missing tee residual roots, and full crowns.That study utilized a dataset of 2278 PRs, consisting 1996 PRs for training and 282 PRs for evaluation, sourced from the Stomatology Hospi of Zhejiang Chinese Medical University (not publicly available due to privacy strictions).It highlighted the potential of CNN-based AI in improving dental diagnost using PRs, leveraging a large dataset for thorough assessment.
As for the papers not previously addressed, namely [30][31][32][33][34][35][36][37][38][39][40], it is evident that a s nificant limitation present in these studies is the lack of comprehensive details regardi the dataset employed.These articles notably omitted crucial information that would o erwise contribute to a more thorough understanding and evaluation of their respecti research methodologies and findings.

Evolution of AI-Diagnostic Tools in Dentistry
A general overview of the AI techniques and architectures employed for various de tal data types is provided as follows:

Convolutional Neural Networks (CNNs)
In dental applications, CNNs are extensively used to analyze image data.They a composed of convolution layers, pooling layers, and fully connected layers.The followi are the primary CNN architectures used: An advanced object detection model that integrates a region proposal network (RP with a CNN.The RPN generates region proposals that are then classified and refined CNN to detect objects and their bounding boxes within an image [7]. Figure 3 shows t Faster R-CNN structure.

ResNet
A deep convolutional neural network architecture that has shown promise in analy ing dental images for various tasks, including caries diagnosis and segmentation, and co sists of multiple layers of convolutional and pooling operations [27].Figure 4 shows t original ResNet-18 architecture.

ResNet
A deep convolutional neural network architecture that has shown promise in analyzing dental images for various tasks, including caries diagnosis and segmentation, and consists of multiple layers of convolutional and pooling operations [27].Figure 4 shows the original ResNet-18 architecture.

NASNetMobile
A lightweight version of the NASNet model designed for mobile and embedded applications.It automatically finds efficient neural network architectures.When combined with Inception V4, it enhances classification performance by generating feature vectors from both networks and using a two-layer fully connected network for final classification [14].Figure 5 shows a mixed model of the Inception V4 + NASNetMobile structure.

YOLOv3
A leading model for object detection and classification tasks.Implemented with the darknet framework, it features 53 layers, including residual blocks from Darknet-53, which enhances its ability to capture intricate image features effectively [24].Figure 6 shows the YOLOv3 network architecture.

NASNetMobile
A lightweight version of the NASNet model designed for mobile and embedded applications.It automatically finds efficient neural network architectures.When combined with Inception V4, it enhances classification performance by generating feature vectors from both networks and using a two-layer fully connected network for final classification [14].Figure 5 shows a mixed model of the Inception V4 + NASNetMobile structure.

NASNetMobile
A lightweight version of the NASNet model designed for mobile and embedded applications.It automatically finds efficient neural network architectures.When combined with Inception V4, it enhances classification performance by generating feature vectors from both networks and using a two-layer fully connected network for final classification [14].Figure 5 shows a mixed model of the Inception V4 + NASNetMobile structure.

YOLOv3
A leading model for object detection and classification tasks.Implemented with the darknet framework, it features 53 layers, including residual blocks from Darknet-53, which enhances its ability to capture intricate image features effectively [24].Figure 6 shows the YOLOv3 network architecture.

YOLOv3
A leading model for object detection and classification tasks.Implemented with the darknet framework, it features 53 layers, including residual blocks from Darknet-53, which enhances its ability to capture intricate image features effectively [24].Figure 6 shows the YOLOv3 network architecture.

NASNetMobile
A lightweight version of the NASNet model designed for mobile and embedded plications.It automatically finds efficient neural network architectures.When comb with Inception V4, it enhances classification performance by generating feature vec from both networks and using a two-layer fully connected network for final classifica [14].Figure 5 shows a mixed model of the Inception V4 + NASNetMobile structure.

YOLOv3
A leading model for object detection and classification tasks.Implemented with darknet framework, it features 53 layers, including residual blocks from Darkne which enhances its ability to capture intricate image features effectively [24].Figu shows the YOLOv3 network architecture.

Relevant Work Experiences
This decade has seen a tremendous development in dental diagnostic tools due to advances in artificial intelligence (AI), machine learning, and automated image analysis.It all started in 2010, when Pedro H. M. Lira and Gilson A. Giraldi published their seminal work on automating the examination of panoramic X-ray images for dental purposes, with an emphasis on feature extraction and teeth segmentation [30].But difficulties like handling teeth that overlapped and the absence of comprehensive dataset information brought attention to the first obstacles.Later research examined the use of artificial neural networks (ANNs) in orthodontic decision-making [31], delving into the subtleties of determining whether extractions are necessary for children of a certain age.Despite attaining an 80% accuracy rate, questions were raised regarding the system's generalizability to various age groups and orthodontic issues.
Years ahead, in pursuit of increasingly advanced diagnostic instruments, Ainas A. ALbahbah et al. ( 2016) employed artificial neural networks (ANNs) and histograms of oriented gradients (HOGs) to identify dental caries [32].Despite showing encouraging results, the study's generalizability and trustworthiness were called into question by the absence of comprehensive dataset information.
Moving into 2017, using image processing techniques, Jufriadif Na'am et al. took on the task of improving the detection of caries between dental X-ray images of adjacent teeth [33].Although the study showed how technology might improve accuracy, it lacked details on datasets and how effective it is in comparison with other methods.
In the same year, Tran Manh Tuan et al.'s consultant system [7], which used fuzzy C-Means clustering and the fuzzy inference system (FIS), pioneered the use of fuzzy rulebased systems for dental diagnostics.The study demonstrated methodology strengths, but encountered constraints in dataset size and specificity.
We now shift the focus to Jae-Hong Lee et al. ( 2018), who evaluated deep CNN models for the identification and diagnosis of dental cavities on peri-apical radiographs.In [8], the landscape moved substantially.While the study achieved strong diagnostic accuracies, particularly for premolars, it had limitations due to its exclusive focus on permanent teeth and the lack of clinical characteristics.
Progressing further in the same year, Gil Jader et al. harnessed the power of deep learning techniques, notably employing the Mask R-CNN architecture, to achieve precise tooth segmentation in panoramic X-ray images marking a significant breakthrough in the field with remarkable accuracy metrics [9].However, despite the groundbreaking nature of their work, the absence of a publishing date and limited dataset details posed significant limitations.
Using dental radiographs and the Faster R-CNN architecture for tooth recognition, in 2019, Tuzoff et al. achieved amazing sensitivity and accuracy that were on par with expert ratings [10].Even with the study's success, its comprehensibility might be improved by a more thorough explanation of the preprocessing stages.
The year 2019 also witnessed Hu Chen et al.'s innovative Faster R-CNN methodology for the detection of teeth in dental periapical films, which demonstrated great recall and accuracy rates and real-world potential [11].Using KNN as a classifier, V. Geethan and K. S. Aprameya, in the same year, elevated machine learning to a new level by concentrating on dental caries diagnosis in dental radiographs [12].The excellent precision and accuracy of the method highlighted machine learning's promise in automated computer-assisted diagnosis systems.
Delving deeper into the same year's discoveries, using a CNN based on the U-net architecture, Vinayahalingam et al. experimented with the automatic detection and segmentation of M3 and IAN in dental panoramic radiographs [13].Although the study was noteworthy for its innovative application of deep learning, it lacked detailed feature extraction standards.
Moving forward to the subsequent year, 2020, an automatic dental X-ray detection technique using a hybrid multi-convolution neural network (CNN) and adaptive histogram equalization was presented by Yaqi Wang et al. [14].Achieving good accuracy, specificity, and AUC, the study indicated a noteworthy step forward, but with a lack of explicit neural network architecture specifications.
Even with these developments, certain research in 2020 remained unclear, including a study that investigated the utilization of deep learning-based artificial intelligence (AI) models for the identification of dental plaque on primary teeth.Notwithstanding improvements in diagnostic precision, the study's stated shortcomings included a small sample size, a narrowly focused analysis, and a lack of knowledge about the CNN architecture [15].
On the other hand, Minyoung Chung et al. (2020) presented a novel technique for identifying individual teeth automatically in dental panoramic X-ray images [16].The method demonstrated a notable improvement in performance over earlier approaches by combining spatial distance regularization (DR) loss with point-by-point localization.
Within the same year, J. Bianchi et al. investigated how artificial intelligence (AI), and specifically deep learning, could revolutionize dental and maxillofacial radiology (DMFR).They emphasized the efficacious outcomes that convolutional neural networks (CNNs) achieved in a range of dental imaging tasks [34].The study highlighted the potential revolution in dental diagnostics while pointing out limitations, such as the lack of huge datasets and creating universally agreed ground truth for AI validation.
In 2020, using panoramic X-ray images, Mircea Paul Muresan et al. took on the task of creating a novel method for automated tooth detection and dental issue categorization [17].The approach showed strengths in precise tooth segmentation and the detection of dental defects by focusing on 14 different dental disorders.Potential drawbacks do, however, include the requirement for increased accuracy.
As we move forward into 2021, Sonavane, A., et al. concentrated on cavity detection with a CNN-based approach, making use of a Kaggle dataset [18].Although the study's greatest accuracy was 71.43%, it was observed that accuracy might be further increased by expanding the dataset.Using optical coherence tomography (OCT) images, Huang and Lee proposed a caries detection study that same year [19].Using the ResNet-152 architecture, they were able to successfully detect caries with a 95.21% success rate.
Continuing with 2021, Alvaro López-Janeiro et al. successfully directed the morphological approach to diagnosis by developing a ML algorithm for enhancing the diagnostic performance for malignant salivary gland tumors [35].Limitations included a small sample size and a lack of external validation.
The landscape of AI in dentistry continued to evolve in 2021 as researchers Jonas Almeida Rodrigues, Joachim Krois, and Falk Schwendicke examined the advancements in AI technology, specifically neural networks, or "deep learning" [36].To address problems such as the lack of generalizability and robustness in many studies, the study stressed the significance of user comprehension and critical assessment of AI applications.
A number of studies examined various aspects of AI in dentistry in 2021.A study examined the efficacy of KNN, SVM, and MLP classifiers in machine learning attribute extraction techniques for the diagnosis of TMD using IT [20].
Nevertheless, the study identified some drawbacks, including a limited sample size, poor generalizability, and a lack of external validation.Another study [37] provided a general review of the many artificial intelligence (AI) approaches utilized in dentistry, but it lacked particular examples of AI applications and comprehensive information on the process and results.
In the same year, Alexey N. Subbotin proposed a smartphone app for dentists to see panoramic X-ray images, with the goal of improving the display of panoramic dentistry photographs using computer technology, fog computing environments, and cloud technologies [38].However, crucial information, such as machine learning classifiers, dataset size, and performance metrics, were missing.
Another study to review for this year is that by Chisako Muramatsu et al.The study investigated a CNN-based method for teeth identification and categorization in dental panoramic radiographs, and it achieved good sensitivity and classification accuracy [21].The study's weaknesses included the small dataset and the absence of molars.
The progress continued in 2022 as ANDAC IMAK et al. proposed a unique technique for detecting caries automatically using periapical images [22].The researchers used a MI-DCNNE with a score-based grouping method to achieve excellent accuracy while recognizing dataset limits.
Furthermore, in 2022, J. Kühnisch et al. investigated artificial intelligence in dental diagnostics, specifically employing CNNs to identify and describe caries from intraoral images [32].The study exhibited an amazing ability to detect cavities while also recognizing the need for further refinement of the technique.
During that same year, Patil et al. [39] explored the applications and pitfalls of AI in diagnosing oral diseases.Their review emphasized the utility of AI in diagnosing dental caries, maxillary sinus diseases, periodontal diseases, salivary gland diseases, TMJ disorders, and oral cancer through clinical data and diagnostic images.They highlighted the immense potential of AI in improving diagnostic accuracy, reducing costs, and minimizing human errors.However, they also pointed out challenges, including the need for larger datasets and the integration of AI into routine clinical practice, which remains a significant hurdle.
Furthermore, in 2022, De Angelis et al. [40] advanced the field further by evaluating the performance of AI diagnostic software named Apox in analyzing panoramic X-rays.Their study demonstrated the AI system's ability to accurately identify dental structures, such as dental formulae, implants, prosthetic crowns, fillings, and root remnants.The results showed high overall sensitivity (0.89) and specificity (0.98), emphasizing the software's reliability in dental diagnostics.However, challenges remained, particularly in detecting radiolucent materials such as fillings and residual roots, which impacted sensitivity.This study underscored the significant progress made in AI-based dental diagnostics while highlighting areas for further improvement to enhance comprehensive diagnostic accuracy.
Also, in the same year, Yassir Edrees Almalki et al. suggested utilizing YOLOv3, a deep learning model, to identify and categorize dental disorders, such as cavities, root canals, dental crowns, and broken-down root canals [24].The study was highly accurate, although it had some drawbacks, including a small sample size and a limited representation of the total population.
During that same year, Abdullah S. AL-Malaise et al. contributed to the development of a method for categorizing dental X-ray images, distinguishing between cavities, fillings, and implants [25].The study produced an accuracy rate of 96.51% using a NASNet model with data augmentation, demonstrating strengths in reaching remarkable accuracy with a small sample size.
Another study in 2022 proposed and evaluated a CNN algorithm for detecting and segmenting MT and MRCs in the maxillary sinus automatically using low-dose and full-dose CBCT [26].Demonstrating high accuracy, the study acknowledged drawbacks, including manual segmentation, an inadequate sample size, and lack of external validation.Additionally, Zhou et al. (2022) sought to enhance traditional CNNs for more accurate identification of dental caries in children on panoramic radiographs, proposing a context-aware CNN that considers information from neighboring teeth [27].
Another work published in the same year focused on the creation of an AI-based system for identifying periodontal bone loss in dental images [28].The study showed great accuracy using deep learning architectures, namely AlexNet and SqueezeNet, with Linear SVM outperforming other classifiers.However, no detailed information regarding the obtained features or EfficientNetB5's classifier was provided, and the evaluation methods and potential constraints were not made obvious, limiting a comprehensive understanding of the study's findings.
The landscape continued to evolve, with Zhu et al. ( 2023) embarking on the development of an AI framework tailored for diagnosing various dental ailments using PRs [29].Utilizing deep CNNs, specifically the BDU-Net and nnU-Net models, the team aimed to enhance the diagnostic efficiency and accuracy of dental pathologies on PRs.The study's diagnostic performance was evaluated based on the AUC, Youden's index, specificity, and sensitivity.The AI framework demonstrated high specificity across multiple dental diseases, such as full crowns, impacted teeth, residual roots, caries, and missing teeth, with AUC values ranging from 0.772 for caries to 0.980 for impacted teeth.That study was significant for its thorough assessment of an AI framework across numerous dental diseases, indicating its promise in dental diagnosis.

Role of X-ray Imaging in Dental Diagnostics
In 2018, Jae-Hong Lee et al. performed a study to assess the efficacy of deep CNN methods for identifying and diagnosing dental cavities on periapical radiographs.Their methodology included the utilization of a pre-trained CNN model, specifically Google Net Inception v3, for initial data processing.Additionally, they employed transfer learning techniques to train their dataset.This investigation focused on leveraging deep learning techniques in dental diagnostics, specifically in the realm of detecting and diagnosing dental caries using X-ray images [8].
Moving into 2019, Vinayahalingam et al. [13] aimed to develop an automated system for detecting and segmenting M3 and IAN on OPGs.Utilizing a CNN based on the U-net architecture, the study showcased the application of deep learning techniques in addressing clinical challenges within dental diagnostics.The emphasis was on automating the recognition and segmentation of dental structures through the utilization of dental X-ray images.
Moving to 2020, Yaqi Wang et al. [14] introduced an automated dental X-ray detection methodology utilizing adaptive histogram equalization and a hybrid multi-convolutional neural network (CNN).This innovative approach addressed the critical role of dental radiograph analysis in clinical diagnostics, where expert interpretation includes tasks such as teeth detection and numbering.The study utilized a dataset comprising 2491 images for training and 138 for testing, categorized into six classes based on different tooth number sequences.To enhance the detection accuracy, three pre-processing techniques were applied: image sharpening and median filtering to remove impulse noise and enhance edges, adaptive histogram equalization to mitigate excessive amplification noise, and a multi-CNN hybrid model for classifying the six different locations of dental slices.The results demonstrated that the accuracy and specificity of the test set exceeded 90%.Additionally, four dentists independently annotated the test dataset.
Moreover, in 2020, Minyoung Chung, Jusang Lee, Sanguk Park, et al. [16] directed their focus to automating the identification and detection of individual teeth in dental panoramic X-ray images.Their proposed technique combined point-by-point localization with spatial distance regularization (DR) loss, offering a potential tool for dental clinics without the need for additional identification algorithms.This research made a valuable contribution to the field of dental diagnostics by introducing a novel method for automating the identification of individual teeth in X-ray images.
In the same year, Mircea Paul Muresan et al. [17] aimed to create a unique technique for automated tooth recognition and dental problem categorization using panoramic X-ray images.Leveraging a deep CNN for semantic segmentation and employing various image processing techniques, the study focused on the accurate identification and categorization of dental abnormalities in panoramic X-ray images.The dataset, composed of panoramic radiographs from different dental clinics, emphasized the utilization of image processing and deep learning methodologies in the field of dental diagnostics.
Fast-forwarding to 2022, Anada Imak et al. [22] proposed a novel method for automatically detecting dental caries using periapical images.Employing a MI-DCNNE model, the study sought to overcome the limitations of manual diagnosis by dentists.Measures such as accuracy, specificity, sensitivity, F1-score, and precision were employed for assessing the model's performance in dental caries detection.The dataset, comprising 340 periapical images from private oral and dental health clinics, underscored the use of deep learning methods in automating the detection of dental caries using X-ray images.
The comprehensive review in Table 1 serves as a brief, yet complete, repository of essential views and conclusions from a collection of papers published between 2010 and 2023, all of which are concerned with AI and machine learning areas of interest.The table, which spans numerous pages, covers key aspects in a structured way, providing a thorough evaluation of the study components.A summary of each category represented in Table 1 is provided below in Figure 7.
Big Data Cogn.Comput.2024, 8, x FOR PEER REVIEW 14 of 37 340 periapical images from private oral and dental health clinics, underscored the use of deep learning methods in automating the detection of dental caries using X-ray images.
The comprehensive review in Table 1 serves as a brief, yet complete, repository of essential views and conclusions from a collection of papers published between 2010 and 2023, all of which are concerned with AI and machine learning areas of interest.The table, which spans numerous pages, covers key aspects in a structured way, providing a thorough evaluation of the study components.A summary of each category represented in Table 1 is provided below in Figure 7.

Impact of AI in Dental Healthcare
The use of artificial intelligence (AI) in dental healthcare has enormous promise, making use of advances in digital imaging and electronic health data.The debate over AI's revolutionary influence on dental practice is lively, but worries remain regarding whether AI will eventually replace the function of oral healthcare professionals.AI excels in simplifying operations and assisting diagnostics due to its ability to use structured knowledge and extract insights from large datasets.However, it falls short of recreating the human brain's sophisticated connections and complicated decision-making skills.A complete knowledge that relies on the skills of dental professionals becomes vital in the field of dental healthcare.Dentists' capacity to perform physical examinations, integrate medical histories, appraise aesthetic outcomes, and engage in meaningful talks remains unrivaled, particularly in situations marked by uncertainty.A machine-led approach to dental treatment is devoid of the human touch that is essential in clinical care.Clinical intuition, intuitive sense, and empathy are critical in providing customized healthcare while maintaining professionalism.While there has been discussion about incorporating empathy into AI algorithms, allowing affective robotics to express artificial emotions, it is critical to emphasize that effective communication between dental healthcare providers and patients involves nonverbal evaluation of hopes, fears, and expectations.These intuitive and spontaneous communication channels, which are essential to the deepest aspects of human-to-human connection, cannot be duplicated in computer language.
The transformative impact of artificial intelligence (AI) and intelligence augmentation (IA) technologies in dental healthcare was explored by Hassani et al. [43].Their study advocates for a future where smart dentistry is characterized by the harmonious blend of AI and IA, leading to improved patient outcomes, enhanced efficiency, and more personalized dental care.This study shows that AI and IA are designed to enhance the intelligence and efficiency of dentists, rather than pose a threat to job opportunities within the dentistry sector.
As a result, AI should be viewed as a supplement, augmenting and occasionally relieving oral healthcare workers to focus on higher-value duties.These may involve synthesizing patient information, enhancing professional contacts, and remaining current on changes in the dental healthcare sector.The progress of dental pedagogy should match with the continual development of AI, providing a harmonic integration that educates future practitioners to harness technology while maintaining the human-centric parts of their practice.

Knowledge Gaps and Future Research Directions
As research studies enhance dental diagnostics with artificial intelligence, our analysis has pinpointed critical knowledge deficits that warrant attention.Primarily, the dataset quality and volume in contemporary research are insufficient, thereby constraining the dependability of AI-driven dental diagnostic tools.Moreover, the validation of AI instruments by dental practitioners in authentic clinical contexts is often wanting, which leads to uncertainties regarding their practical implementation.Additionally, there is a shortfall in the clarity and transparency of AI methodologies within the dental sector, which can impede the professional uptake of these advancements.Research concerning AI-facilitated dental diagnostics is also restricted to a handful of dental conditions, limiting its breadth and prospective influence.
AI's progress in dental diagnostics necessitates addressing numerous crucial aspects to assure the technology's usefulness, dependability, and acceptability in clinical practice.First, increasing dataset quality and quantity is critical.This entails investing in the acquisition and curation of big, diverse, and high-quality information, which will allow AI models to learn more accurately and generalize more broadly across populations.Collaborative clinical validation is also required, which encourages cooperation between AI developers and dental practitioners to undertake complete clinical trials.This real-world testing and iterative improvement will improve the practical applicability and dependability of AI technologies while also instilling trust in dental practitioners.Furthermore, demystifying AI technology through educational programs and tools designed specifically for dental professionals would increase comprehension and enable adoption into daily practice.Expanding the scope of dental problems researched by widening research efforts to encompass both common and unusual illnesses would increase the adaptability of AI diagnostic tools.Interdisciplinary collaboration among AI specialists, dental researchers, practitioners, and other stakeholders is required to create creative solutions and a comprehensive strategy for incorporating AI into dentistry.Establishing explicit ethical principles and legal frameworks will guarantee that AI is developed and deployed responsibly, addressing concerns like data protection and responsibility.Finally, building user-friendly AI diagnostic tools with simple interfaces and seamless interaction with existing dental practice systems would help dental professionals to accept and use them more easily.Focusing on these future areas might help the field of AI-enhanced dentistry diagnostics to overcome present constraints and pave the way for more effective, efficient, and widely accepted AI applications, ultimately leading to significant advancements in dental care.

Conclusions
In conclusion, this exhaustive review highlights the significant progress and challenges associated with integrating artificial intelligence (AI) within the realm of dental medicine.AI's advent has been a game-changer for interpreting dental radiographs and identifying oral health issues.Nonetheless, various hurdles impede the full exploitation of AI's capabilities in enhancing dental care.A key obstacle is the limited availability and variable quality of data, upon which the success of AI models is heavily contingent.Specifically, there is a lack of data on rare dental diseases, complex cases, high-quality annotated radiographs, and procedural variations.Data security and privacy also play a crucial role, given the sensitive nature of dental records and imagery.
AI algorithm transparency and intelligibility remain considerable barriers, impacting the accuracy and reliability of AI systems due to inconsistent data labeling and non-uniform annotations.Therefore, the development of clear, accountable AI frameworks is of the essence.Documenting AI's journey in dentistry, this review acknowledges initial endeavors and subsequent breakthroughs with advanced learning algorithms, emphasizing AI's promise in improving diagnostic precision and efficiency in oral health care.It highlights that artificial intelligence (AI) is a supplement to the valuable human aspects that dental professionals afford.Instead, AI should be used to help dental practitioners to make clinical decisions.Highlighting knowledge gaps, this study emphasizes the need for large, highquality datasets, effective validation of AI tools in practical settings, more transparency, and greater AI use across a variety of dental difficulties.
Furthermore, this paper points out the limits of AI in catching nonverbal signs, highlighting the need for human empathy and intuition in dental treatment.While AI can evaluate data and aid in diagnoses with great precision, it cannot read delicate nonverbal signs, such as a patient's body language, facial expressions, and emotional responses.These indicators are critical for recognizing patient discomfort, anxiety, and pain, which have a substantial impact on the quality of care.Human dentists may respond to nonverbal cues with empathy and change their approach accordingly, fostering trust and providing a more individualized treatment experience.Addressing these flaws is critical for exploiting the totality of AI's capabilities in the dentistry industry, ensuring that AI enhances, rather than reduces, the quality of patient care.

Figure 2 .
Figure 2. Summary of the procedure for identifying dental diseases.
depicts a process flow diagram that serves as a visual guide for the diagnostic procedures of dental ailments.

Figure 2 .
Figure 2. Summary of the procedure for identifying dental diseases.

Figure 2 .
Figure 2. Summary of the procedure for identifying dental diseases.

Table 1 .
Comparative analysis of papers published in 2010-2023 on dentistry, where N/A means not mentioned.