The Role and Applications of Artificial Intelligence in Dental Implant Planning

1. Introduction

Artificial intelligence (AI) has become increasingly prevalent in various fields of medicine, revolutionizing healthcare delivery and patient outcomes. From diagnostic imaging to treatment planning, AI technologies have demonstrated remarkable capabilities in enhancing clinical decision-making and improving efficiency [1]. In dentistry, AI is emerging as a valuable tool for enhancing various aspects of patient care, including implant planning—a critical component of dental implantology that demands precision and meticulous planning. By leveraging AI algorithms and machine learning techniques, clinicians can analyze complex datasets and optimize treatment strategies for individual patients [2]. Implant planning involves the evaluation of patient anatomy, bone density, and other factors to determine the optimal position, size, and angle of dental implants. Traditionally, this process has relied heavily on the expertise and experience of dental professionals, often involving manual measurements and subjective assessments. [3,4]

However, the integration of AI into implant planning brings forth a new era of precision and efficiency. AI algorithms can analyze vast amounts of patient data, including radiographic images, three-dimensional scans, and clinical records, to assist clinicians in making evidence-based decisions regarding implant placement [5]. Moreover, AI-powered software can provide predictive modeling and simulation capabilities, allowing clinicians to visualize the expected outcomes of different treatment approaches before initiating the procedure. This not only enhances treatment planning but also enables personalized and patient-specific interventions [6]. Despite the potential benefits, the widespread adoption of AI in implant planning raises several ethical, legal, and practical considerations. Issues such as data privacy, algorithm transparency, and liability pose significant challenges that must be addressed to ensure the responsible and ethical use of AI technologies in dentistry [7].

In this article, we aim to explore the role and applications of AI in implant planning, examining the current state of the art. Additionally, we will discuss the ethical and legal implications associated with the integration of AI into clinical practice, providing insights into the opportunities and challenges of this rapidly evolving field. Through a comprehensive review of the literature, we seek to provide dental professionals with a deeper understanding of the potential of AI in implant planning and its implications for patient care, clinical workflow, and professional responsibilities.

2. Materials and Methods

This review was conducted in accordance with the guidelines of PRISMA [8]. Based on the PICO criteria, a search strategy was developed, and an electronic search was conducted. The PICO question was formulated as follows: “What are the current advantages and diagnostic aids provided by AI in the field of implant planning?”. A comprehensive literature search was conducted on PubMed and Scopus databases to identify relevant studies related to the role and applications of artificial intelligence (AI) in implant planning. In addition, Google Scholar was reviewed. The search was performed using a combination of keywords and Medical Subject Headings (MeSH) terms, including: ((Artificial Intelligence [Mesh] OR (AI) OR (machine learning) OR (deep learning)) AND ((Implant Planning [Mesh] OR (implantology) OR (implant treatment plan)). The search was limited to articles published in English from inception to the present. (Table 1)

Further manual exploration of the reference lists of all full-text articles and relevant reviews identified from the electronic search was also conducted. Additionally, manual searches were carried out in the following journals: Journal of Prosthodontic Research, Journal of Prosthetic Dentistry, Clinical Oral Implants Research, International Journal of Oral Maxillofacial Implants, Clinical Implant Dentistry and Related Research, Implant Dentistry, and Journal of Implantology. The inclusion criteria for the studies were defined as studies of any level of evidence, with the exception of expert opinion. Additionally, only articles published in English and within the last 5 years were considered. The exclusion criteria were set to omit review articles and letters to editors, as well as animal studies. Moreover, studies were excluded if the full text was unavailable. Two independent reviewers screened the titles and abstracts of the retrieved articles to identify potentially relevant studies. Full-text articles were then assessed for eligibility based on the inclusion criteria outlined above. From the selected articles, we gathered the following information: the names of the author(s), the publication year, the country of origin, and the study design. We also noted the total number of patients or datasets, details on the training and validation datasets, and the test datasets used in the studies.

Additionally, we documented the study's objective, the application of artificial intelligence, and the conclusions and outcomes reported by the authors. The quality of included studies was evaluated using appropriate quality assessment tools, such as the Cochrane Risk of Bias Tool. Studies were assessed for risk of bias, methodological limitations, and potential sources of bias. (Table 2)

The main characteristics of the included studies, such as the study design, the AI techniques applied, and the published findings, were summarized using descriptive statistics. The research' conclusions were examined using a qualitative synthesis, which sought to identify recurring themes, patterns, and areas in which the investigations concurred or differed. Ethical standards and criteria for systematic reviews and meta-analyses were followed in the conduct of this review. Since the study utilized the analysis of publicly available data from other published studies, no ethical approval was needed. The omission of publications written in languages other than English and the potential for publication bias are two potential weaknesses of this review.

Furthermore, the overall conclusions and interpretations could have been impacted by the caliber and diversity of the included studies. Overall, the approach used in this study was to methodically locate, assess, and compile the existing data regarding the function and uses of AI in implant planning, offering insightful information about the state of the field at the moment and guiding future avenues for investigation.

3. Results

On April 1, 2024, the systematic search came to an end. After screening 256 article titles, 30 abstracts were chosen for additional examination. Twenty articles were then examined in their entirety to see if they satisfied the inclusion requirements. Following additional review, 6 articles were eliminated for the following reasons:

• Not a study in the field of AI application in implant planning (n = 3);
• Full text not available (n = 1);
• Missing information on AI technology (n = 2).

A total of 14 full-text papers were included for data extraction after two more articles that met the inclusion criteria were found through manual searching. [9,10,11,12,13,14,15,16,17,18,19,20,21,22,23] All data collected, Study Design, Aim of the Study, AI Application, Outcome or Conclusions, are collected in Table 3

4. Discussion

Diagnostic procedures are a fundamental pillar in dental practice. An accurate diagnosis and detection of anatomic structures is the first step towards effective and personalized treatment. [24,25] In the field of oral implant rehabilitation, is essential, to avoid intraoperative and postoperative complications, to plan the implant placement away from neural structures. [26,27]

Therefore, accurate recognition of neural structures is of paramount importance. The first and most crucial step in planning mandibular implant treatment involves identifying the position of the inferior alveolar canal. [28] From the review we conducted, 4 scientific articles propose the identification of the mandibular nerve through the use of AI in implant diagnostic evaluation. All four studies describe the processes through which novel automated processes based on CNNs are developed to segment mandibular neural structures. Shuo Yang et al. evaluated the recognition of the inferior alveolar nerve on two-dimensional panoramic images. The authors used machine learning on 1366 panoramic images, demonstrating that this method achieved high performance for IAC segmentation in panoramic images under different visibilities. Additionally, the authors demonstrated that the comparison between automated segmentation and manual annotation showed that the IAC position was highly consistent between the two segmentation approaches, with a matching degree close to 85%. Another study assessed the outcomes of a deep convolutional neural network on both two-dimensional and three-dimensional images obtained through machine learning from a dataset of 49,094 images acquired from CBCT scans. The authors conclude that while 3D U-Net demonstrated significantly superior results compared to 2D Net in automated canal nerve detection, deep learning will significantly enhance the efficiency of treatment planning. The other two studies evaluate the segmentation of neural structures on three-dimensional CBCT images. Oliveira-Santos N et al. in their study mentioned the use of an AI-driven tool that provided accurate segmentation of the mandibular canal, even in the presence of anatomical variations such as an anterior loop. Thus, the presently validated dedicated AI tool may aid clinicians in automating the segmentation of neurovascular canals and their anatomical variations. It may significantly contribute to presurgical planning for dental implant placement, especially in the interforaminal region. Also, in the methodology described by Jindanil T. et al. satisfactory results were achieved. Through machine learning of 200 CBCT scans, the authors concluded that the software autonomously recognizes the mandibular nerves. All four studies, despite the differences in computer methods, lead to the same results, namely that automated segmentation of the mandibular incisive canal on CBCT scans was proven to be accurate, time-efficient, and highly consistent, serving pre-surgical planning.

Two studies, Kurt Bayrakdar S. et al. and Mangano F. et al., instead evaluate models that more comprehensively detect and automatically segment various anatomical structures such as nerves, sinuses, bones, and missing teeth. Kurt Bayrakdar S. et al. also include measurements related to bone thickness and height. Additionally, Mangano F. et al. incorporate another crucial aspect into digital implant planning, namely the matching between CBCT segmentation and intraoral digital scanning, also achieved effectively in an automated manner.Among researchers in the field of machine learning, there is a debate about what should be the sufficient dataset size to create a proper learning curve for software. In all the retrospective cohort studies we have selected, we did not find homogeneity in the datasets used. Among the studies that used three-dimensional volumes of CBCT, the dataset varied from a minimum of 43 to a maximum of 200. Among the studies that evaluated the software based on two-dimensional images derived from panoramic X-rays or CBCT sections, the minimum dataset was 316 images, while the study with the largest dataset had 16,272 images. Only the research conducted by Roongruangsilp P. et al. investigated the learning curve of the developed AI for dental implant planning, suggesting that regarding automated learning using images, the number of each image category used in AI development is positively related to AI interpretation. All authors agree that fifty images are the minimum image requirement. The primary stability achieved after implant insertion is crucial for promoting osseointegration. [29] This stability is influenced by the bone drilling protocol and the hardness of the bone. [30] A very interesting study explores the possibility of establishing a correct implant protocol supported by AI software. Takahiko S. et al. demonstrated that artificial intelligence, after training on 960 images (taken from 2D sections of CBCT), can provide an effective method of predicting drilling protocols. A more recent study also assesses the clinical reliability of an AI-assisted implant planning software program with an in vitro model. Chen Z. et al. obtained very satisfactory results and conclude by stating that the use of AI-supported software can effectively program the ideal implant position through self-learning. Additionally, they evaluate bone density, concluding that higher bone density led to increased implant deviations, emphasizing that it is one of the fundamental parameters to consider during implant planning. One article, instead, was selected concerning the anamnestic evaluations to be performed on the patient before implant surgery. Lyakhov et al. introduces an artificial intelligence system designed to analyze patients' statistical factors to predict the success of dental implant survival accurately. By utilizing a digitized database of clinical cases of osseointegration and an optimally designed neural network architecture tailored to these factors, a neural network system with a testing accuracy of 94.48% was achieved.

Peri-implant site underpreparation is commonly used by implantologists, especially in cases of very soft bone. The degree of underpreparation is generally dictated by the clinician's experience. [31] In implantology, the use of radiographic stents with landmarks helps to guarantee precise surgical planning. [32] This method involves the use of a radiographic stent, which is a personalized guide for the individual based on the patient's radiographic data. [33] The radiographic stent contains radiographic landmarks, which are identifiable reference points on the panoramic radiograph or CBCT scan. Before or during the surgical procedure, the radiographic stent is placed on the patient's panoramic radiograph or CBCT scan to ensure precise alignment of the landmarks with the patient's anatomical structures. This allows the surgeon to accurately locate the position and angulation of dental implants based on pre-operative planning. [34] Alsomali D. et al. developed a model that automatically localizes the position of radiographic stent markers in CBCT scans. However, the authors did not achieve reliable results. They clearly emphasize that the use of only axial images for training an AI program for localization of guttapercha markers is not sufficient to provide an accurate AI model performance. This result is consistent with the findings of the study by Hyunjung K.G. et al., which highlights how the use of three-dimensional models showed significantly better results than 2D models. In the realm of medical science, diagnostic procedures are pivotal for the accurate detection of any anatomical structures and also to aid in accurate diagnosis and subsequent any further treatment plans. Recently, the use of convolutional neural networks (CNNs) has proven to provide excellent performance in the field of 3D images analysis.

Firstly, the most crucial step in planning an implant in the maxillary region is identifying the anatomical structures and assessing the bone quality and quantity and any adjacent vital structures ex: maxillary sinus, nasal cavity, infraorbital nerve. [35] Hence, from the review we conducted 2 studies were proposed the use of CNN in the detection of vital anatomical structures in the maxillary region through the use of AI in implant diagnostic evaluation. Both the two studies describe the processes through which novel automated processes based on CNNs are developed to segment the maxillary sinus and bone. Nermin M. et al. in her study developed and validated a novel automated CNN-based methodology for the segmentation of maxillary sinus using CBCT images. The author used in her study a dataset of 264 sinuses images that were acquired from 2 CBCT devices. The author also had demonstrated that this method achieved high performance for the detection of segmentation of the maxillary in CBCT images and had proposed CNN model provided a time-efficient, precise, and consistent automatic segmentation which could allow an accurate generation of 3D models for diagnosis and virtual treatment planning to the comparison of manual segmentation of the maxillary sinus on CBCT images which the author proposed as time- consuming and dependent on the practitioner’s experience with high inter- and intra-observer variability.

Another study assessed the outcomes of the use of CNNs in the maxillary region Cavalcante F. R. et al the author evaluated a total of 141 CBCT scans and concluded the AI-driven segmentation was 116 times faster than the manual segmentation yet, although the use of manual segmentation showed slightly better performance, the novel CNN-based tool also provided a highly accurate segmentation of the maxillary alveolar bone and its crestal contour consuming 116 times less than the manual approach. In another study within of its limitation relative to its small size a total of 43 CBCT datasets. It had investigated the use of Artificial Intelligence solution to Auto-segmentation in the edentulous mandibular bone for Implant planning area using Artificial Intelligence Adel Moufti M et al. For this purpose, the author concluded that the use of segmentation of the edentulous spans on CBCT images was successfully conducted by machine learning with good accuracy compared to manual segmentation. It had achieved a good accuracy (>90%) in segmenting unliteral cases, which represent the majority of patients with missing teeth it had achieved a good level of accuracy in segmenting edentulous bone areas compared to the human investigators. This automation of bone assessment on CBCT images has the potential to significantly reduce the time and associated cost of implant treatment.

Analyzing the potential legal implications of personal data protection in AI studies is also interesting. Health data is protected with a high level of protection, such as within the European Union [Regulation (EU) 2016/679 of the European Parliament and of the Council of 27 April 2016 on the protection of natural persons with regard to the processing of personal data and on the free movement of such data and repealing Directive 95/46/EC (General Data Protection Regulation)], requiring an impact assessment of the potential use of the data prior to processing. It is thus essential to make the data to be used anonymous, not only by checking whether the patient can be identified in the document to be evaluated, but also by removing the metadata that may be present in a radiology examination that contains personal data. [36] This is essential when using external applications, as clinical data security may not be guaranteed. Ethics committees approving AI studies must also be aware of the potential leakage of personal data that may occur in this type of study and may even suggest additional security systems to researchers. [37]

5. Conclusions

Although the number of studies indicates a growing interest in the application of AI in medicine, studies conducted on the use of modern artificial intelligence technologies for implant planning highlight the need for high-quality training data and the lack of standardization in protocols.

However, even with different methods and datasets, automated recognition and segmentation methods of anatomical structures such as bones, nerves, teeth, and maxillary sinuses were proven to be accurate, time-efficient, and highly consistent. New studies are needed to evaluate implant success rates and drilling protocol assessments. The use of these methods conducted by machine learning can significantly contribute to efficient treatment planning, and the development of larger datasets can further enhance AI application performance.