Diagnostic evaluation of the relationship between impacted mandibular third molars and the mandibular canal in panoramic radiographs shared via WhatsApp

Abstract

Objective

To compare the diagnostic accuracy of computer and smartphone screens for assessing the radiographic proximity between impacted mandibular third molars and the mandibular canal.

Methods

Four examiners evaluated 121 digital panoramic radiographs for the presence of proximity between third molars and the mandibular canal, as well as for specific radiographic signs indicative of this relationship. Assessments were performed on a computer screen and on their personal smartphone screens; for the latter, images were shared via WhatsApp. Diagnostic performance was analyzed using accuracy (Ac), sensitivity (Se), specificity (Sp), and positive and negative predictive values (PPV and NPV). Differences in the frequency of correct and incorrect responses between methods were tested using the McNemar test, as were the proportions of correct and incorrect classifications for each radiographic sign.

Results

Examiners correctly classified true positives and true negatives in 86.5% of cases using the computer monitor and 85.1% using the smartphone. Comparable diagnostic performance was observed for computer monitor (Se = 0.975; Sp = 0.758; Ac = 0.865) and smartphone (Se = 0.942; Sp = 0.762; Ac = 0.851) assessments. No statistically significant differences were found between methods for any radiographic sign (p > 0.05).

Conclusion

Smartphones demonstrated diagnostic accuracy comparable to that of computer screens for evaluating the proximity between impacted mandibular third molars and the mandibular canal on panoramic radiographs.

Introduction

In dentistry, high diagnostic accuracy is essential, and obtaining a second opinion can be helpful for complex cases. Consequently, sharing radiographic images via instant messaging on smartphones is common among health professionals. Sarode et al. [1] found that 98% of dentists regularly used instant messaging apps to send clinical photographs and radiographs for second opinions. WhatsApp is one of the most widely used apps, allowing rapid message transmission across nearly all mobile operating systems [2].

Wani et al. [3] first evaluated WhatsApp for healthcare communication in plastic and reconstructive surgery, finding it effective, inexpensive, fast, and easy to use. Subsequent studies confirmed its effectiveness for medical information exchange and reliability in analyzing radiographic images for diagnostics [4–9]. The European General Data Protection Regulation [10] mandates specific rules to protect patients’ rights, but no global guidelines exist for WhatsApp’s use in clinical practice [11, 12].

Extraction of the mandibular third molar is a common oral surgery [13, 14]. Preoperative evaluation of the relationship between an impacted third molar and the mandibular canal is crucial. Panoramic radiography is the preferred imaging technique due to its availability, ease of use, low cost, and low radiation dose [15]. Despite some overlapping and distortion, it is recommended for evaluating the surgical site and identifying perioperative risks [13, 16, 17]. Specific high-risk signs include interruption of the radiopaque line of the mandibular canal wall, deviation or narrowing of the mandibular canal, and a radiolucent area over the molar roots, all indicating close anatomical proximity and a higher risk of inferior alveolar nerve injury [17–22].

The European Society of Radiology [23] notes that viewing radiographic images on mobile devices is not recommended for primary interpretation; it facilitates sharing for second opinions. The accuracy of diagnosing the relationship between the included mandibular third molars and the mandibular canal using WhatsApp remains unclear. No studies specifically address this issue. Given the growing use of mobile image sharing for better diagnoses, further research is warranted.

Materials and methods

This cross-sectional observational study was conducted at the Diagnostic Imaging Laboratory of the School of Dentistry at the Federal University of Rio Grande do Sul (UFRGS). The study adhered to the guidelines outlined in Resolution 466/2012 of the National Health Council, which governs research involving human subjects. Prior to initiation, the study received approval from the Research Ethics Committee of UFRGS (CEP-UFRGS), under the substantiated CAAE No. 56710022.0.0000.5347.

Case selection

A sample size of at least 112 mandibular third molars was calculated to test for a minimum difference of 0.15 points between the AUC (area under the curve) of the methods under analysis, with 80% power and 5% significance. The calculation utilized the ‘power.roc.test’ function from the pROC package in R. To ensure robustness and the largest required sample size, we assumed A1 = A2 = 0 (mean test scores in both positive and negative conditions equal), B1 = B2 = 1 (test scores in both conditions have the same variability), and RN = ra = 0.5 (equal prevalence of negative and positive conditions).

This study utilized 121 panoramic radiographs, each measuring 2864 × 1348 pixels with a resolution of 336 dpi. These images were obtained using a flat-panel digital sensor and a Carestream CS8100 dental X-ray machine from the image bank of the Digital Image Processing Laboratory (LAPID) at the UFRGS School of Dentistry. The study included panoramic radiographs with diagnostic quality that clearly showed mandibular third molars with complete rhizogenesis. The sample was divided into two groups. Only one mandibular third molar per radiograph was included. The first group consisted of 60 images of fully formed, impacted mandibular third molars whose root apices were in proximity to the mandibular canal. The second group comprised 61 images in which the molars showed no close anatomical relationship with the mandibular canal. The reference standard was established by consensus through the analysis of radiographs by three professors from the UFRGS School of Dentistry: two from Oral Radiology (MBV and NAA) and one from Oral and Maxillofacial Surgery and Traumatology (DP). Each case was evaluated for signs of direct contact between the third molar and the mandibular canal, and for specific predictive radiographic signs of proximity, including: (a) interruption of the canal wall’s radiopaque line, (b) radiolucent band or area over the third molar roots, (c) deflection of the canal around the root apices, and (d) narrowing of the roots.

Image evaluation

All students enrolled in the Integrated Residency Program in Oral Health at the School of Dentistry, with an emphasis on Oral and Maxillofacial Surgery and Traumatology, were invited to participate in the study. The examiner group consisted of six residents with experience in interpreting panoramic radiographs. The Coordination of the Integrated Residency Program in Oral Health at the School of Dentistry was informed of the residents’ voluntary involvement in this research. Students who agreed to participate were instructed to read and sign the informed consent form (ICF). A total of 121 radiographs comprised the sample, and these were randomized for each of the two analysis stages using the Research Randomizer application (https://www.randomizer.org).

In M1, the radiographs and their corresponding questions were compiled in a Google Form, with each image followed by two sequential case-related questions. Examiners viewed and answered the questions on a computer monitor and submitted the completed form electronically.

In M2, a separate Google Form containing only the questions was emailed to the examiners. The radiographs, identified by case and tooth numbers, were shared via WhatsApp in JPG format and were accessed and analyzed on smartphone screens. The examiners subsequently completed the questions and submitted the form electronically.

Both methods used panoramic radiographs with a resolution of 2864 × 1348 pixels, 336 DPI, and a file size of approximately 1.16 megabytes (1,219,622 bytes), shared in JPG format. In each visualization method, the examiners answered two questions for every case analyzed. The first asked them to determine whether there was direct contact between the included molar and the mandibular canal, using a five-point Likert scale: YES, certainly; YES, probably; I DON’T KNOW; NO, probably not; and NO, definitely not. For the second question, if the examiner responded “yes,” they were instructed to select one or more predictive radiographic signs of proximity.

Statistical analysis

All analyses were conducted using the R language [24]. Accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated to compare evaluations with the gold standard. McNemar’s χ² test was used to compare discordance between two dichotomous variables, particularly to compare sensitivity and specificity between diagnostic methods in paired data. Sensitivities were compared based on evaluation proportions of each method in positive diagnosis cases, while specificities were compared among negative diagnosis cases using McNemar’s test [25]. Cohen’s Kappa coefficient measured agreement between the evaluations made by the same evaluator on two different occasions. Agreement levels were classified as low (0 < K < 0.20), reasonable (0.21 < K < 0.40), moderate (0.41 < K < 0.60), substantial (0.61 < K < 0.80), and perfect (0.81 < K < 1.0), with K < 0 indicating no agreement [26]. A significance level of α = 0.05 was applied to all analyses, and p-values at or above this threshold were interpreted as evidence to retain the null hypothesis.

Results

Table 1 presents the intra-rater agreement analysis using Cohen’s Kappa statistics. Evaluators completed the questionnaire using method 1 twice at different times, and agreement between their responses was assessed. The original 5-level response scale collapsed into three categories: Yes (definitely yes and probably yes), No (definitely not and probably no), and don’t know. Kappa (K) values ranged from 0.54 to 0.88. Analyses of the four highest agreement values were used in the statistical analysis.

Table 1.

Analysis of intra-rater agreement in method 1 at two different times according to Cohen’s Kappa statistic (K) for each evaluator

Evaluator	K	CI 95%		p	Category of concordance [26]
1	0.55	0.41	0.68	< 0.001	Moderate
2	0.88	0.80	0.97	< 0.001	Perfect
3	0.87	0.78	0.96	< 0.001	Perfect
4	0.69	0.54	0.85	< 0.001	Substancial
5	0.54	0.39	0.69	< 0.001	Moderate
6	0.72	0.59	0.84	< 0.001	Substancial

Table 2 displays precision measures of evaluations compared to the gold standard for both methods. Accuracy rates were 86.5% using computer screen and 85.1% for smartphone (using WhatsApp). Computer visualization outperformed smartphones in accuracy, sensitivity, PPV, and NPV, while smartphones showed higher specificity. Despite this, most metric values were similar between the two methods. Both methods demonstrated high accuracy, sensitivity, specificity, PPV, and NPV, ranging from 0.758 to 0.975, indicating good evaluation accuracy compared to the gold standard. Table 3 presents the frequencies of “yes” and “no” responses for each method and compares the proportions using the McNemar test. This statistical approach allows for a comparison of sensitivity and specificity values shown in Table 2.

Table 2.

Measures of accuracy of the evaluations in comparison with the gold standard, considering all evaluators in the two methods used in the study

Measurement	Method 1	Method 2
Accuracy	0.865	0.851
CI 95%	(0.832, 0.895)	(0.816, 0.882)
Sensitivity	0.975	0.942
Specificity	0.758	0.762
Positive predictive value	0.798	0.796
Negative predictive value	0.969	0.930
N	483	484

Table 3.

Comparison of frequencies of “yes” and “no” responses between the two methods in reference to the gold standard and significance of the McNemar test

Gold Standard	Methods		Method 2			X² McNemar (p)
			No	Yes	Total
Yes	Method 1	No	6	0	6	0.0133
		Yes	8	225	233
		Total	14	225	239
No	Method 1	No	173	12	185	1
		Yes	13	46	59
		Total	186	58	244
Both	Method 1	No	179	12	191	0.1637
		Yes	21	271	292
		Total	200	283	483

The McNemar test indicates that the difference in sensitivities was significant (p = 0.0133), while the difference in specificities was not (p = 1). This means that computer visualization was significantly more sensitive than smartphone using WhatsApp. Table 3 also displays the overall frequencies of “yes” and “no” responses. The McNemar test result was not significant (p > 0.05), suggesting no significant difference in the proportion of responses between Method 1 and Method 2. Thus, the change in response proportions between the methods was not statistically significant.

Figure 1 illustrates the distribution of evaluators’ responses on the Likert scale relative to the gold standard for each method, allowing comparison of response proportions. For the gold standard “NO,” most responses were “NO, probably no direct contact” in both computer screen (53.7%) and smartphone using WhatsApp (56.6%). For the gold standard “YES,” the predominant response was “YES, there is probably direct contact,” with proportions of 87.5% and 85.8% for computer screen and smartphone using WhatsApp, respectively. These results suggest a tendency for evaluators to favor responses indicating probability over certainty.

Fig. 1.

Distribution of evaluators’ responses in relation to the gold standard by method on a scale from 1 (No, there is definitely no direct contact) to 5 (Yes, there is definitely direct contact)

The study proceeded with the evaluation of four radiographic signs identified by the evaluators: (1) “Radiolucent area over the roots of the third molar,” (2) " Narrowing of the roots of the third molar,” (3) “Deviation of the mandibular canal,” and (4) “Discontinuity of the cortical bone of the mandibular canal.” Table 4 presents the frequencies for each radiographic sign across the two methods, along with the total number of cases and the significance values from McNemar’s test. Overall, these results suggest no statistically significant difference between the proportions of correct and incorrect responses for any radiographic sign between the two methods.

Table 4.

Tables of frequencies of hits and misses between the two methods for each radiographic signal and significance of the comparison according to McNemar’s test

Radiographic sign	Method 1	Method 2			Total		Teste de McNemar (p)
		Correct	Incorrect
1) Radiolucent area	Correct	102	5		107		0.7237
	Incorrect	3	2		5
	Total	105	7		112
2) Narrowing of the roots	Correct	30	7		37		0.6276
	Incorrect	10	29		39
	Total	40	36		76
3) Deviation of the canal	Correct	16	4		20		1
	Incorrect	5	15		20
	Total	21	19		40
4) Discontinuity of the cortical bone	Correct	31	13		44		0.3827
	Incorrect	8	24		32
	Total	39	37		76

Discussion

Smartphones are the most widely used devices for sharing and viewing images [27]. They have high-speed processors, large memory, high-resolution screens, fast internet, and free access to apps like WhatsApp. These features facilitate quick consultations for second opinions, providing fast and effective access to specialized knowledge. In Oral Radiology, sharing radiographs instantly can speed up diagnosis and treatment [28]. However, the reliability of radiographic images shared via WhatsApp is uncertain, as smartphone screens may not be ideal for viewing. This study assessed smartphone use for radiographic diagnosis and found an 85% accuracy rate in evaluating the mandibular third molar and mandibular canal relationship in panoramic images shared via WhatsApp. These findings are consistent with other studies on diagnostic accuracy using smartphone screens [27, 29–32].

Recent research indicates that the file format of periapical radiographs does not affect diagnostic accuracy for external and internal root resorption and vertical root fractures [33, 34]. In this study, the same file format was used for both methods. Madi et al. [30, 31] demonstrated WhatsApp’s excellent reliability for diagnosing jaw fractures and pathological conditions, with 99% accuracy and agreement (k) above 0.9. Comparing the two methods, computer screens were slightly superior overall. Both methods had high sensitivity values, indicating that panoramic radiographs are more reliable for confirming proximity between analyzed structures than excluding it, regardless of the screen used. Despite the limitations of two-dimensional examinations, computer screens had significantly higher sensitivity than smartphones, suggesting computer screens offer better conditions for identifying proximity between roots and the mandibular canal. Evaluators in this study identified multiple markers, contributing to high sensitivity values. According to Atieh [17], finding two or more radiographic markers improves sensitivity in predicting true proximity relationships.

This study identified radiographic markers predicting the proximity between the mandibular third molar and the mandibular canal. Evaluators most frequently noted the radiolucent area on the roots in both methods, consistent with previous findings [18]. Delamare et al. [35] found direct root-canal contact in 85% of cases with this marker. The radiolucent area often indicates a lingual position of the mandibular canal, associated with a higher risk of inferior alveolar nerve injury [16, 36] and higher incidence of post-extraction sensorineural deficits [14]. Comparing the proportions of correct answers between the two methods showed no statistically significant difference for any radiographic sign, suggesting equal difficulty in identifying markers across methods. Method 1’s greater sensitivity and frequent identification of multiple signs per case imply that identifying multiple markers is more crucial than the type of marker itself, aligning with previous research [17–22].

This study had limitations, including the lack of prior evaluator training and absence of strict ambient light control, to mimic clinical conditions. Computer monitors were uniform, and smartphones had standardized specifications to ensure diagnostic compatibility.

A major concern with sharing radiographs via WhatsApp is the potential exposure of sensitive patient information. Despite end-to-end encryption, WhatsApp lacks specific features for protecting medical data. Without global guidelines for such applications in healthcare, professionals must ensure data protection. The increasing use of WhatsApp in medicine highlights the need for regulation and monitoring of sensitive data sharing by competent authorities.

The diagnostic accuracy of panoramic radiographs shared via WhatsApp in smartphones for evaluating the relationship between impacted mandibular third molars and the mandibular canal was similar to that of computer screens. These findings align with current literature, highlighting smartphones as a reliable tool for dentists seeking cases discussion and second opinions.

Funding

The Article Processing Charge (APC) for the publication of this research was funded by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - Brasil (CAPES) (ROR identifier: 00x0ma614).