Selective Cone-Beam CT Referral for Mandibular Third Molars Based on Panoramic Risk Signs: A Retrospective Cross-Sectional Diagnostic Study

1. Introduction

The extraction of mandibular third molars is one of the most common procedures in oral and maxillofacial surgery and carries a potential risk of inferior alveolar nerve (IAN) injury, with reported incidences reaching up to 8.4% in the literature [1,2,3]. Proper surgical planning requires an accurate assessment of the anatomical relationship between the roots and the inferior alveolar canal (IAC) in order to minimise the risk of complications.

Panoramic radiography (OPG) is typically the initial imaging technique due to its wide availability, low cost, and reduced radiation dose [4,5]. However, being a two-dimensional modality, it may present distortions and superimpositions that hinder interpretation, particularly in cases where the roots lie close to the IAC [4,5,6,7,8]. Cone-beam computed tomography (CBCT) offers precise three-dimensional evaluation, but its indiscriminate use results in increased radiation exposure, unnecessary costs, and the risk of trivialising its indication [9,10,11,12].

In many clinical settings, access to CBCT has increased substantially, including in general dental practices. This broader availability facilitates immediate imaging decisions at the point of care, but may also contribute to variability in prescription patterns and potential overuse when clear and standardised criteria are lacking. Consequently, selective CBCT referral strategies should be applicable to real-world clinical settings, where imaging decisions are often made by clinicians with different levels of experience. Therefore, this study assessed the performance of a pragmatic, guideline-aligned selective CBCT referral algorithm and examined how examiner experience affects its applicability.

The aim of this study was to evaluate interobserver agreement in the identification of panoramic radiographic risk signs, analyse their association with root–IAC contact confirmed by cone-beam computed tomography, and retrospectively assess a selective imaging protocol.

2. Materials and Methods

This retrospective cross-sectional diagnostic study was conducted in a university-based oral surgery setting and reported in accordance with the STARD guidelines. The study protocol received approval from the University’s Research Ethics Committee, and all participants signed informed consent in accordance with the Declaration of Helsinki. Radiographic records obtained between 2021 and 2025 were retrospectively reviewed.

2.1. Selection Criteria

Patients aged 18 to 90 years were included if they presented mandibular third molars with complete root formation and had both preoperative panoramic radiography (OPG) and cone-beam computed tomography (CBCT) available.

Exclusion criteria were: low-quality imaging; local alterations preventing accurate interpretation (cysts, periapical lesions, mandibular fractures, or osteosynthesis hardware); pregnancy; history of head and neck radiotherapy; ongoing oncologic treatment; or refusal to undergo radiographic studies.

2.2. Sample Size and Unit of Analysis

Sample size was calculated at the tooth level, because the primary outcome of the study was CBCT-confirmed root–IAC contact for each mandibular third molar. Assuming a 95% confidence level, a conservative expected proportion of 50% to maximise sample size requirements, and an accepted precision of approximately ±11%, a minimum sample of 80 mandibular third molars was required.

The final sample included 62 patients and 86 mandibular third molars, therefore exceeding the minimum estimated sample size. Each mandibular third molar was independently assessed by four examiners, generating 344 individual radiographic evaluations. Diagnostic accuracy analyses, regression models, ROC curve analysis, and retrospective application of the selective CBCT indication protocol were performed at the tooth level.

Interobserver agreement analyses were based on the individual examiner assessments of each tooth. When both mandibular third molars from the same patient were included, they were analysed as separate anatomical units because each tooth had an independent spatial relationship with the IAC. The potential non-independence of bilateral molars from the same patient was considered when interpreting the findings and is acknowledged as a limitation.

Four examiners with different levels of clinical experience assessed the panoramic radiographs: a first-year postgraduate student, a second-year postgraduate student, a third-year postgraduate student, and a faculty oral surgeon with more than 10 years of clinical experience. Each examiner recorded the presence or absence of the radiographic signs described by Rood and Shehab [1], including cortical interruption, root darkening, curved or deflected roots, root narrowing, canal deviation or narrowing, bifid apices, and “island-shaped” apices (Figure 1). To reduce measurement bias, all panoramic radiographs were independently assessed using predefined radiographic criteria.

CBCT imaging (Planmeca ProMax 3D Classic®) was used as the reference standard to confirm root–IAC contact and to measure the minimum distance (in millimetres) between the root and the IAC.

2.3. Statistical Analysis

Statistical significance was set at P < 0.05. Numerical variables were described using the mean and standard deviation or the median and interquartile range (Q1–Q3), according to their distribution.

Interobserver agreement was assessed using the kappa coefficient (κ) with 95% confidence intervals.

To analyse the association between radiographic signs and root–IAC contact, a two-stage modelling approach was applied:

• Logistic regression to estimate the probability of root–IAC contact.
• Linear regression was used to assess the relationship between radiographic findings and root–IAC distance among mandibular third molars without CBCT-confirmed contact.

The optimal cutoff for the number of panoramic signs predicting complete root–IAC contact was determined using Youden’s index. Diagnostic performance was evaluated through sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and ROC curve analysis.

The association between root position relative to the IAC and the radiographic signs was examined using independent logistic regression models.

Data processing and statistical analyses were performed using R software version 4.3.2 [13].

For the retrospective analysis, CBCT indication was operationally defined as the presence of cortical interruption or ≥2 panoramic risk signs. Examiner experience and medico-legal considerations were incorporated as clinical modifiers in the proposed decision algorithm, but not as additional numerical criteria for the retrospective diagnostic performance analysis.

3. Results

3.1. Sample Characteristics and Imaging Findings

The characteristics of the study sample, imaging assessments, examiner profile, and main CBCT findings are summarized in Table 1.

3.2. Distribution of Radiographic Signs

Among the 86 mandibular third molars evaluated, the most frequent radiographic signs on panoramic imaging were cortical interruption and root darkening (30.6% and 37.5%, respectively). Curved or deflected roots were observed in 18–21% of cases. Root narrowing, canal narrowing, and canal deviation were less common findings (<10%). Bifid apices and “island-shaped” apices were identified only anecdotally. These results are presented in Supplementary Figure S1.

3.3. Interobserver Agreement

Interobserver agreement for the identification of the different radiographic signs was generally low among most examiners, with kappa coefficients close to zero and wide confidence intervals.

The highest agreement was observed between Examiners 3 and 4, particularly for root darkening (κ [95% CI] = 0.64 [0.48–0.79]), curved or deflected roots (κ [95% CI] = 0.59 [0.39–0.80]), and root narrowing (κ [95% CI] = 0.64 [0.33–0.96]), showing moderate to substantial levels of agreement.

Interobserver agreement between the two most experienced examiners is shown in Table 2. The highest agreement was observed for root darkening, curved or deflected roots, and root narrowing, with κ values ranging from 0.59 to 0.64 in the total sample. Agreement was lower for interruption of the cortical outline, canal deviation, and canal narrowing. Kappa values could not be estimated for island-shaped apex because this sign showed no variability between the two examiners.

3.4. Association Between Radiographic Signs and Root–IAC Contact

Cortical interruption was the only sign significantly associated with the presence of root–IAC contact (OR = 3.46; 95% CI: 1.23–10.83).

In contrast, among mandibular third molars without root–IAC contact, cortical interruption was not associated with either greater or lesser distance to the canal (β [95% CI] = 0.10 [–0.78 to 0.99]; p = 0.816). The optimal cutoff for predicting complete root–IAC contact based on the number of radiographic signs was ≥1 sign present. CBCT-confirmed root–IAC contact was more frequent among mandibular third molars presenting at least one panoramic risk sign than among those without radiographic signs. Diagnostic accuracy estimates were calculated using CBCT-confirmed contact as the reference standard. Figure 2 displays the corresponding ROC curve using this cutoff: sensitivity, 0.90 (95% CI: 0.77–0.97); specificity, 0.38 (95% CI: 0.24–0.54); PPV, 0.59 (95% CI: 0.46–0.71); and NPV, 0.80 (95% CI: 0.56–0.94). Figure 3 illustrates the relationship between the cumulative number of panoramic signs and CBCT-confirmed root–IAC contact or distance. Although ≥1 sign maximized sensitivity according to Youden’s index, the final selective protocol prioritized clinical applicability by combining the strongest individual predictor, cortical interruption, with the presence of multiple radiographic signs.

3.5. Comparison Between OPG and CBCT

The distances measured between the root and the IAC were significantly greater on CBCT than on OPG (P < 0.001), with a moderate effect size (Cohen’s d = 0.72). The intraclass correlation coefficient indicated moderate agreement between the two methods (ICC = 0.66). Bland–Altman analysis revealed a systematic tendency for OPG to underestimate the root–IAC distance compared with CBCT.

3.6. Root Position Relative to the IAC

Logistic regression analysis identified curved or deflected roots as the only radiographic feature significantly associated with IAC position. The presence of this sign was associated with an 83% reduction in the odds of a lingual IAC location compared with molars without root curvature. These results are presented in Supplementary Figure S2.

3.7. Application of the Selective Protocol

The retrospective application of the selective CBCT indication protocol would have recommended CBCT in 33/86 mandibular third molars (38.4%). Among the 38 molars with CBCT-confirmed root–IAC contact, 33 fulfilled the protocol criteria, whereas five contacts would not have been identified by the protocol. No molars without CBCT-confirmed root–IAC contact fulfilled the protocol criteria. Accordingly, the protocol showed a sensitivity of 86.8%, specificity of 100.0%, PPV of 100.0%, and NPV of 90.6%. These findings should be interpreted as an internal retrospective performance estimate rather than external validation of the protocol. The diagnostic performance of the selective CBCT indication protocol is shown in Table 3. The decision algorithm is shown in Figure 4.

4. Discussion

This study evaluated the reliability of panoramic radiographic signs for predicting contact between mandibular third molar roots and the IAC, using CBCT as the reference standard.

Our results show that cortical interruption was the only panoramic sign significantly associated with CBCT-confirmed root–IAC contact, supporting its diagnostic relevance. However, interobserver agreement varied across signs and was generally higher among more experienced examiners. These findings are consistent with classic studies [1,2,3] as well as more recent investigations [4,5], which also identified this sign as a reliable predictor of surgical risk. In our analysis, the presence of two or more panoramic signs on OPG increased the probability of contact, reinforcing the usefulness of a combined radiographic assessment. Similar results have been reported in studies comparing the diagnostic performance of OPG and CBCT in evaluating root–IAC relationships [10,11,12].

The low kappa values observed among less experienced examiners should not be interpreted as an indication for routine CBCT in all cases. Rather, they reflect the limited reproducibility of subtle panoramic signs when interpreted by junior operators, a finding that may be partly influenced by the low prevalence of several signs and the prevalence-dependence of kappa statistics. Clinically, this supports the need for structured training, use of standardised visual criteria, and senior review in borderline cases before dismissing the need for CBCT.

The higher concordance observed was between Examiners 3 and 4, reflecting that interobserver agreement was influenced by the level of clinical experience. This finding highlights the importance of specialised training and the learning curve in radiographic interpretation, consistent with studies that have shown that diagnostic reliability depends on experience [5,6].

Regarding linear measurements, OPG systematically underestimated the root–IAC distance compared with CBCT. This pattern, previously reported [7,8], confirms that while OPG is useful as an initial examination, it has inherent limitations related to its two-dimensional nature. CBCT provides more precise three-dimensional anatomical information and may improve surgical planning in selected high-risk cases, although its routine use has not consistently been shown to reduce postoperative neurosensory complications [11,12,13,14,15,16,17].

The retrospective application of the selective CBCT protocol identified 33 of 38 CBCT-confirmed root–IAC contacts (86.8%) while recommending CBCT in fewer than 40% of mandibular third molars. The protocol showed high specificity and PPV in this retrospective sample, as no false-positive CBCT indications were observed. However, because the protocol was derived and tested within the same dataset, these estimates may overestimate its true clinical performance. Therefore, the proposed algorithm should be considered a pragmatic referral framework requiring prospective external validation rather than a definitive prediction rule. These findings suggest that a rational use of CBCT based on the presence of panoramic risk signs, modulated by examiner experience and surgical complexity, may optimise the balance between diagnostic confidence and radiation exposure, in accordance with justification and optimisation principles [9,16]. In this regard, Sanmartí-García et al. [18] suggested that CBCT may be useful in selected high-risk cases, although current evidence does not support routine CBCT for all mandibular third molars. Recent evidence also supports a selective rather than indiscriminate use of cone-beam computed tomography in mandibular third molar surgery. Ruiz-Roca et al. reported that the combination of root darkening and interruption of the cortical line on panoramic radiography increased the probability of IAC-related sensory alteration, while Robbins et al. concluded that adding CBCT to panoramic imaging does not routinely reduce nerve injury rates when used indiscriminately [19,20,21,22].

4.1. Strengths and Limitations

The main strengths of this study include the participation of examiners with different levels of clinical experience, the use of CBCT as the reference standard, an a priori sample size calculation, and the application of a clinically oriented selective CBCT indication protocol. In addition, the analysis of interobserver agreement provides relevant information on the influence of examiner experience in the interpretation of panoramic radiographic signs.

However, several limitations should be acknowledged. First, this was a single-centre study with a limited sample size, which may restrict the generalisability of the findings. Second, although the primary analysis was performed at the tooth level, some patients contributed more than one mandibular third molar; therefore, a possible clustering effect of bilateral molars within the same patient cannot be completely excluded. Third, individual anatomical determinants, such as root morphology, canal cortication, depth of impaction, angulation, and buccolingual canal position, were not systematically incorporated into multivariable models. Finally, although clinical follow-up was performed, the study was not designed to establish a standardised longitudinal correlation between radiographic findings and postoperative neurosensory outcomes.

Future multicentre prospective studies with larger samples, external validation cohorts, and more comprehensive anatomical covariates are needed to confirm the clinical applicability of the proposed selective CBCT indication protocol.

4.2. Selective CBCT Indication

Based on these findings, CBCT should be considered when cortical interruption is present or when two or more panoramic risk signs are identified. In cases with a single non-cortical sign, the decision should be individualised according to surgical complexity, examiner experience and the need for medico-legal documentation. In the absence of panoramic risk signs, OPG may be sufficient for initial surgical planning. This approach reserves CBCT for higher-risk cases and aligns with radiological justification and optimisation principles.

4.3. Clinical Implications

Panoramic radiography remains an appropriate first-line imaging modality for the preoperative assessment of mandibular third molars. However, its limitations should be recognised, particularly in cases with suspected proximity between the roots and the IAC. In this study, cortical interruption was the panoramic sign most strongly associated with CBCT-confirmed root–IAC contact, supporting its use as a key criterion for selective CBCT referral.

The findings also highlight the influence of clinical experience on radiographic interpretation. Less-experienced operators may benefit from structured training, standardised visual criteria, and senior review in borderline cases. Therefore, CBCT should be considered when cortical interruption or multiple high-risk panoramic signs are present, especially in complex surgical cases or when medico-legal documentation is required.

This selective approach may improve diagnostic confidence while avoiding routine CBCT use, thereby reducing unnecessary radiation exposure and healthcare costs. It provides a pragmatic framework for balancing patient safety, surgical planning, and radiological justification in mandibular third molar surgery.

5. Conclusion

In conclusion, cortical interruption was the most consistent panoramic indicator of CBCT-confirmed root–IAC contact. However, the generally low interobserver agreement, particularly among less experienced examiners, highlights the limitations of relying solely on panoramic interpretation. A guideline-aligned selective approach, incorporating cortical interruption, multiple signs, operator experience, and clinical complexity, may support more consistent CBCT referral decisions while avoiding routine imaging.