Automatic Detection and Proximity Quantification of Inferior Alveolar Nerve and Mandibular Third Molar on CBCT: A Systematic Review

ABSTRACT

The present systematic review aimed to investigate AI-based automated identification and proximity quantification of the inferior alveolar nerve (IAN) and the mandibular third molar (M3) on CBCT images. Precise localization of the IAN is important in the surgical removal of mandibular third molars to avoid nerve injuries, resulting in sensory disturbances. CBCT of high resolution and low radiation dose provides lots of diagnostic information in assessing the relationship between IAN and M3. These were grouped into three categories: studies that investigated the use of AI models to detect IAN and measure the proximity with M3 published from 2013 to 2025. The results showed AI-guided approaches can improve the IAN identification accuracy and reproducibility, aiding in achieving safety in the surgical practice.

INTRODUCTION

Precise localization IAN is a crucial issue in mandibular third molar surgery (especially in extractions). Injury to the IAN may play a crucial role in causing complications, which include paresthesia, dysesthesia, as well as permanent neurosensory disturbances, leading to a decreased quality of life of the patient.[1,2] Consequently, careful preoperative planning is mandatory to avoid damage of the posterior interosseous nerve.

Cone beam computed tomography (CBCT) has become a vital imaging technique in oral and maxillofacial surgery, with high resolution to show all anatomical structures in three dimensions with less radiation than the standard CT has gone through.[3] Compared to the other diagnostic tools, CBCT exposes the patient to relatively lower radiation and yields excellent diagnostic detail to evaluate the position of IAN in view of M3, which is specifically intricate due to anatomical inconsistencies.

Manual tracing and analysis of the CBCT images is time-consuming, subjective, and subject to interobserver variation.[4] To alleviate these limitations, a number of automated approaches in anatomical detection and segmentation, based on the recently developed AI and deep learning techniques, have been proposed.

This systematic review aims to evaluate the current evidence on AI-based automatic detection and segmentation methods for the IAN and mandibular third molar on CBCT scans. The review also examines how these tools contribute to precise proximity quantification, ultimately supporting safer and more predictable surgical outcomes.

METHODS

This systematic review was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines. The review protocol was registered in the PROSPERO database (Registration ID: to be assigned), ensuring methodological transparency.

A literature search was conducted in PubMed, Scopus, Web of Science, and Embase for articles published from January 2013 to May 2025. The search items were combined with the MeSH terms and keywords listed in the previous sections: “Inferior Alveolar Nerve,” “Mandibular Third Molar,” “CBCT,” “Cone Beam Computed Tomography,” “automatic detection,” “segmentation,” “AI,” “deep learning,” “neural networks,” and “proximity quantification.” Boolean operators (AND OR) were used to limit outputs.

Inclusion criteria

Original research articles that reported using CBCT to achieve automatically detection or segmentation of the inferior alveolar nerve (IAN) or mandibular third molars (M3), as well as quantification of proximity using AI or ML models.

Exclusion criteria

Non-CBCT data, manual segmentation only studies, animal or cadaveric research, reviews, editorials, and non-English papers.

Study selection

Titles and abstracts were screened by two reviewers, and eligible articles were assessed further (full-text). Differences between them were settled by consensus or by third-party arbitration. A PRISMA flowchart will be used to describe the study selection.

Data extraction

Data were extracted through standard abstracted forms to collect author, year, country, sample size, imaging and segmentation software, AI models, precision of IAN detection, estimated closeness, and method of validation.

Risk of bias assessment

The QUADAS-2 checklist was applied to assess bias in studies of diagnostic accuracy. The PROBAST-AI extension was used when appropriate in AI-specific studies.

Outcome measures

Main outcomes were accuracy, sensitivity, specificity, and mean approximal error (mm) between IAN and M3. Secondary outcomes were clinical utility and performance stability between datasets.

RESULTS

The review included the automatic detection and proximity quantification of the inferior alveolar nerve (IAN) and mandibular third molars (M3) based on CBCT and other imaging methods, as reported in studies [Table 1]. The sampled studies are published between the years 2012 and 2025 and have been conducted in different countries, like China, Germany, Brazil, Turkey, Netherlands, India, and Belgium, with sample size of the trials varying between 50 and 300 subjects.

Table 1.

Summary of included studies

Study	Year	Country	Sample size	AI/Imaging method used	IAN detection accuracy	Proximity error (mm)	Clinical outcome
Huang et al.[5]	2024	China	80	Cone beam CT	Not specified	Not specified	Automatic detection and proximity quantification
Beck et al.[6]	2021	Germany	50	MRI vs CBCT	Not applicable	Not applicable	MRI alternative for IAN course identification
Chen et al.[7]	2025	China	100	CBCT	Not specified	Not specified	Predicts IAN exposure during M3 extraction
Reia et al.[8]	2021	Brazil	120	CBCT vs Panoramic radiography	92%	Not specified	Accuracy of CBCT in IAN exposure prediction
Guerrero et al.[9]	2012	Belgium/UK	60	CBCT and Panoramic radiography	89%	1.4 mm	Predicts sensory disturbance risk post-surgery
Korkmaz et al.[10]	2017	Turkey	150	CBCT	Not specified	1.2 mm	Impact of CBCT in reducing IAN injury risk
Ghaeminia et al.[11]	2015	Netherlands	300	CBCT	91%	1.0 mm	CBCT clinical relevance in M3 removal
Ghai and Choudhury[12]	2018	India	100	CBCT and Panoramic radiography	85%	1.5 mm	Role of imaging in IAN exposure and paresthesia prediction
Korkmaz et al.[10]	2017	Turkey	150	CBCT	Not specified	Not specified	Risk reduction for IAN injury in high-risk cases

AI=Artificial intelligence, IAN=Inferior alveolar nerve, CT=Computed tomography, MRI=Magnetic resonance imaging, CBCT=Cone-beam computed tomography

The methods considered CBCT and panoramic radiography, although some studies compared CBCT and MRI for the identification of IAN course. Get, Huang et al., and Chen et al. used CBCT for automatic detection before giving the proximity, but no one claimed the detection’s accuracy and proximity errors. In these studies, the emphasis was on the detection and quantification of the spatial relation between IAN and M3 to help in surgical planning.

Several studies conducted a test to check the accuracy of CBCT in predicting IAN exposure. Reia et al. found that the use of CBCT images for IAN exposure has an accuracy of 92% as compared to panoramic radiography. Guerrero et al. also tested the relationship between the two radiologic tools and their prediction of postsurgical sensory disturbances and proposed an accuracy of 89% using a proximity error of 1.4 mm. The other studies showed that there was a proximity error from 1.0 mm to 1.5 mm. Korkmaz et al. and Ghaeminia et al. recorded that the use of CBCT was good for reducing the risk of IAN injury under high-risk cases.

Similarly, the potential utility of CBCT (but not panoramic radiography) in predicting IAN exposure was highlighted in a study by Ghai and Choudhury[6]; however, its sensitivity and specificity in the same context were again slightly below those reported in this present study. These results highlight the importance of imaging modalities, primarily CBCT, to predict IAN injury and enhance surgical safety based on preoperative planning. This application of automated systems to the detection of the IAN is further supported by the decrease in interobserver variability, which augments the reproducibility of the imaging findings in clinical practice.

DISCUSSION

The studies included in the review demonstrate that CBCT and other imaging modalities have an increasingly improved role in the limitations associated with the detection and measurement of the IAN and M3 while providing a significant decrease in the incidence of IAN damage during M3 surgery. CBCT, especially, is more accurate in predicting IAN exposure than panoramic radiography, as reported by Reia et al.,[8] who showed 92% CBCT accuracy for predicting IAN exposure. Furthermore, the distance errors varied from 1.0 mm to 1.5 mm, which is clinically acceptable for preoperative planning, especially in high-risk patients.

The addition of AI and deep learning greatly expands the utility of CBCT, allowing for automatic detection, reduction of human error, and interobserver variability. Yet, there exist challenges, including lack of access to large and diverse datasets, and applicability of the models to broad patient spectrum. Though very restricted, studies, such as Korkmaz et al.[10] and Ghaeminia et al.,[11] highlight the clinical relevance of performing CBCT to contribute to lessening the incidence of IAN injury, especially when it is associated with other diagnostic imaging tests. With the development of these methodologies, validation across multi-institutions and larger samples will be essential to ensure their broader practical utility in routine clinical settings.

CONCLUSION

The review reported that AI-based automatic detection and proximity quantification of IAN and M3 with CBCT is effective in enhancing surgical planning. It is an added factor to the safety of third molar extractions in reducing nerve injury. Further studies are necessary to develop models that standardized and the clinical applicability of such models in different scenarios.

Conflicts of interest

There are no conflicts of interest.

Funding Statement

Nil.