All 41 participants were included in the final analysis. Sixteen patients were included in the case series. The other 14 patients were excluded due to neither the IAN nor LN being clinically suspected of being involved or no iatrogenic traumatic cause (Fig. 2). These patients were included in the control group, together with the 11 healthy volunteers (total n = 25).

The cases had a total of 18 injuries: 9 to the LN and 9 to the IAN. One patient had injuries to both lingual nerves and another to both inferior alveolar nerves. All other patients in the case series had an injury to a single nerve.

The case series consisted of 10 females and 6 males, and the control group of 16 females and 9 males. There was no significant difference in sex between cases and controls (p = 0.92).

Age in the case series varied between 16 and 62 years, with a mean age of 40 years. In the control group, age varied between 13 and 83 years, with a mean age of 51 years. The difference in mean age between both groups was not significant (p = 0.66).

Clinical data could be extracted from the medical files of all 16 patients in the case series. Neuropathic pain was present in nine patients. The others experienced neurosensory disturbances without them being described as painful.

Clinical Sunderland classifications based on NST included eight class I, two class II, two class III, two class IV, three class V, and one undetermined injury (examples given in Fig. 3).

Illustrative case studies demonstrating varying degrees of nerve injury. Coronal plane 3D CRANI images. A Bilateral normal inferior alveolar nerve (IAN). B Grade I degree of right-sided IAN injury. Homogeneously increased T2 signal with no change in caliber. C Grade II degree of right-sided IAN injury. Homogeneously increased T2 signal and mild nerve thickening. D Grade III degree of right-sided IAN injury. Homogeneously increased T2 signal for nerve and moderate to marked nerve thickening, perineural fibrosis. E Grade IV injury of right IAN injury. Heterogeneously increased T2 signal and focal enlargement in otherwise continuous nerve. F Left, an end-bulb neuroma and transection of the left lingual nerve compatible with a class V injury

Iatrogenic causes of trauma were implant placement (n = 2), tooth extraction (n = 8), xanthoma curettage (n = 1), bilateral sagittal split osteotomy (BSSO; n = 1), BSSO + genioplasty (n = 1), open reduction internal fixation (ORIF; n = 1), and iatrogenic undefined (n = 2).

Inter-rater agreement for injury detected on MRN, nerve thickness, nerve signal intensity (SI nerve), and MRN Sunderland classification score was substantial, moderate, good, and moderate, respectively (Table 2).

Intra-rater agreement for observer one for injury detected on MRN, nerve thickness, SI nerve, and MRN Sunderland classification score was moderate, moderate, excellent, and moderate, respectively (Table 2). For observer two, they were almost perfect, good, good, and substantial, respectively.

Overall, compared to NST, MRN had a sensitivity (true-positive rate) of 38.2% and specificity (true-negative rate) of 93.5% (Table 3). Positive likelihood ratio, negative likelihood ratio, positive predictive value, and negative predictive value were 5.9, 0.66, 46, and 91.3, respectively.

When differentiated by clinical Sunderland class, both groups had a specificity of 93.5%. Sensitivity differed between both, with the low clinical Sunderland class group having a sensitivity of 19.1% and the high clinical Sunderland class having a sensitivity of 83.3%. Positive likelihood ratios in the low and high clinical Sunderland class groups were 2.96 and 12.89, respectively. The same tendency was seen when differentiated for both nerves. For the LN, the global sensitivity, specificity, and positive likelihood ratio were 48.6%, 96.5%, and 13.95. Differentiated according to clinical Sunderland class, specificity remained the same in both groups but sensitivity differed. Sensitivity and the positive likelihood ratio in the low clinical Sunderland class group were zero because of the absence of any true-positive results. In the high clinical Sunderland class group, sensitivity was 81.8%, with a positive likelihood ratio of 23.45.

The sensitivity, specificity, and positive likelihood ratio for the IAN group were 28.2%, 90.7%, and 3.02. Specificity remained the same when differentiated by clinical Sunderland class. In the presence of a low Sunderland class, the sensitivity and positive likelihood ratios were 25.7% and 2.76. For the higher classes, these values were 100% and 10.72, respectively.

There was a moderate, positive correlation between clinical and MRN Sunderland classification scores. The Spearman correlation coefficient based on the constructed contingency table was 0.53 (p = 0.005; Table 4).

The prediction model using aSNR, aNMCNR, and nerve thickness to predict the presence of injury had an area under the receiver operating characteristic curve of 0.78 (p =  < 0.05), with an F-score of 0.19 and an accuracy rate of 0.89. The permutation feature importance test showed the following levels of importance for the different variables: 408.03 for aNMCNR (p =  < 0.05), 293.33 for aSNR (p =  < 0.05), and 28.50 for nerve thickness (p =  < 0.05). Additional receiver operating characteristic analyses of aSNR in combination with nerve thickness and aNMCNR in combination with nerve thickness were performed due to the multicollinearity between aSNR and aNMCNR. The area under the receiver operating characteristic curve, F-score, and accuracy rate were 0.73, 0.01, and 0.89, respectively (p =  < 0.05), for the model using aSNR and 0.77, 0.96, and 0.92 (p =  < 0.05) for the model using aNMCNR.

Differences in aSNR, aNMCR, and nerve thickness between healthy and injured nerves are shown in Table 5. A significant difference in mean nerve thickness was found for the overall dataset but not for both nerves separately. For aSNR and aNMCNR, a significant difference was found for the overall dataset and both nerves separately.