Diagnosis of trigeminal neuralgia based on plain skull radiography using convolutional neural network

Diagnosis of trigeminal neuralgia based on plain skull radiography using convolutional neural network | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Diagnosis of trigeminal neuralgia based on plain skull radiography using convolutional neural network Jung Ho Han, So Young Ji, Myung Ju Kim, Ji Eyon Kwon, Jin Byung Park, and 7 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4775021/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 29 May, 2025 Read the published version in Scientific Reports → Version 1 posted 10 You are reading this latest preprint version Abstract This study aimed to determine whether trigeminal neuralgia can be diagnosed using convolutional neural networks (CNNs) based on plain X-ray skull images. To this end, 166 skull images of patients aged > 16 years with trigeminal neuralgia diagnoses were compiled into a labeled trigeminal neuralgia dataset and 498 skull images of patients with unruptured intracranial aneurysms were compiled into a labeled control dataset. The images were partitioned into training, validation, and test datasets in a 6:2:2 ratio using random permutation. The accuracy and area under the receiver-operating characteristic (AUROC) curve were used to evaluate the classifier performance. Gradient-weighted class activation mapping was employed to identify the focal areas of attention. External validation was performed using a dataset obtained from another institution. We observed an overall accuracy of 87.2%, sensitivity and specificity of 0.72 and 0.91, respectively, and AUROC of 0.90 on the test dataset. In most cases, trigeminal neuralgia was predicted by observing the sphenoid body and clivus. The overall accuracy on the external test dataset was 71.0%, indicating the promise of deep learning-based models in distinguishing between X-ray skull images of patients with trigeminal neuralgia and control individuals. This is expected to serve as a useful screening tool after further development. Health sciences/Anatomy Health sciences/Neurology Health sciences/Signs and symptoms Trigeminal neuralgia Deep learning Convolutional neural networks Skull X-ray Figures Figure 1 Figure 2 Figure 3 Figure 4 INTRODUCTION Trigeminal neuralgia is characterized by recurrent, unilateral, short bursts of electric shock-like pain affecting one or more branches of the trigeminal nerve. 1 The most widely accepted theory regarding the cause of trigeminal neuralgia is neurovascular compression (NVC). 2 However, recent studies have challenged this notion, suggesting that NVC may not be a sufficient or necessary condition for trigeminal neuralgia development, 3 prompting a closer examination of additional predisposing factors. Higher prevalence of trigeminal neuralgia in females and Caucasian individuals further points to the potential involvement of other etiological factors. 4–6 Before the acceptance of NVC as a cause of trigeminal neuralgia during the 1960s and 70s, structural changes in the skull were considered mechanical factors in trigeminal neuralgia etiology. 7,8 Early studies using plain skull X-ray imaging identified morphological differences associated with trigeminal neuralgia, such as bony angles at the ridge of the petrous bone. 9,10,11 With advances in imaging technology, recent studies have explored the role of neuroanatomical structures in trigeminal neuralgia etiology, revealing smaller posterior fossa and cistern volumes and increased crowding in patients with trigeminal neuralgia. 12,13 These findings suggest that there are additional structural features that distinguish patients with trigeminal neuralgia from normal individuals. Although the prevailing consensus on trigeminal neuralgia etiology leans towards NVC, the practical utility of discerning subtle differences in plain skull X-ray images with the naked eye has diminished over time. However, with recent breakthroughs in image analysis using deep learning, renewed exploration of these claims is warranted. This technology enables novel re-evaluation of the significance of structural variations in trigeminal neuralgia, potentially providing valuable insights that were previously too challenging to ascertain. In the realm of medical imaging, deep learning, particularly convolutional neural networks (CNN), has exhibited great promise in recognizing features that may elude human observation. 14 Therefore, in this study, we evaluated the applicability of trigeminal neuralgia in distinguishing plain skull X-ray images using CNN. MATERIALS AND METHODS Study enrollment The study design was reviewed and approved by the Institutional Review Boards of Seoul National University Bundang Hospital (B-1910-572-101) and Ajou University Medical Center (AJIRB-MED-MDB-22-234), and it adhered to the recommendations of the Declaration of Helsinki for biomedical research involving human subjects. This study complied with all IRB ethical regulations. Informed consents were waived by the Committee due to the retrospective and deidentified nature of the data. During the collection of lateral skull radiographs for investigation, the following inclusion criteria were used: 1) patients aged over 16 years, and 2) patients with lateral skull radiographs without any trace of previous craniofacial surgeries or any truncation of the image. We enrolled clinically established patients with trigeminal neuralgia experiencing paroxysmal unilateral orofacial pain distributed along the trigeminal territory which should be triggered by typical maneuvers. 15 To identify eligible patients with trigeminal neuralgia, we used trigeminal neuralgia registry and medical records. In aggregate, 277 patients were diagnosed with trigeminal neuralgia between January 2013 and June 2020. As a control group, we selected patients with unruptured intracranial aneurysms who underwent clipping surgery because most patients underwent skull radiography for preoperative planning. In aggregate, 3,045 patients were identified using a clinical data warehouse query system. Among them, 2,706 underwent skull radiography and were ready for review. After reviewing all medical records and skull radiographs, 166 patients with trigeminal neuralgia and 1,702 control group patients were retained as qualifying study subjects with appropriate skull X-ray images. To control confounding effects, we performed three-fold matching by age and sex. 16 As a result, 664 patients (166 trigeminal neuralgias and 498 controls) were enrolled in this study. All subjects in the control group were confirmed to not have undergone Rany procedures or surgeries related to facial pain and exhibited no history of related prescriptions. Each group was partitioned into training, validation, and test datasets in a 6:2:2 ratio using random permutation. Generalized performance of the CNN was evaluated on an external test dataset consisting of 100 patients (50 trigeminal neuralgias and 50 controls) from other institutes, in the same manner as described above. The patient selection process is illustrated in Fig. 1 . Convolutional neural network classifier We used TensorFlow (version 2.0.0, https://www.tensorflow.org/ ) as a framework to train and evaluate the neural network model in skull radiograph classification. Pre-trained ‘Densely Connected Convolutional Network’ (DenseNet-121) 17 with ImageNet dataset 18 and ChesXNet dataset 19 were imported for initial weight configuration. Originally, the pre-trained DenseNet-121 was designed to identify 1,000 (ImageNet) and 14 (ChesXNet) labels using 224 × 224 images. However, as we required a binominal classifier, the top layers, consisting of 1,000 and 14 fully connected nodes, respectively, were replaced with layers that included two nodes. The original softmax activation function was replaced by the sigmoid function \(\:(\:f\left(x\right)=\:\frac{1}{1+{e}^{-x}}\:)\) , with 7,039,554 parameters. The loss function was set to be a binary cross-entropy ( \(\:-{y}_{true}\text{log}f\left(x\right)\:-\left(1-{y}_{true}\right)\text{log}(1-f\left(x\right)\:)\) ). Various optimizers (SGD, RMSprop, Adagrad, Adadelta, Adam, Adamax, and Nadam), with initial learning rates ranging between 10 −6 and 10 −3 , were tested to achieve best model performance. Preprocessing, model training, and evaluation Considering the modest number of data points in the training dataset, image augmentation was performed on it in two ways. First, contrast-limited adaptive histogram equalization was applied to enhance particular characteristics by improving image contrast. 20 Second, rotation within 15°, translation within 10%, transposition within 10%, or brightness alteration within 20% were randomly applied in combination (Fig. 2 ). Moreover, to account for class imbalance (1:3), the control group was augmented by a factor of 14 to 4,172 images and the trigeminal neuralgia group by a factor of 20 to 2,000 images. For the validation dataset, only contrast-limited adaptive histogram equalization was applied to produce two iterations for each label. The test dataset was used without any modifications. Several performance indicators were calculated from the confusion matrix, which consisted of the numbers of true positives (TP, predicts TN as TN), true negatives (TN, predicts control as control), false positives (FP, predicts control as TN), and false negatives (FN, predicts TN as control). The accuracy, precision, and recall were defined to be \(\:\frac{TP+TN}{TP+FP+TN+FN}\) , \(\:\frac{TP}{TP+FP}\) , and \(\:\frac{TP}{TP+FN}\) , respectively. The F1-score was defined as the harmonic mean of precision and recall. The area under the receiver-operating characteristic curve (AUROC) was calculated using the probability produced by the sigmoid output for each label. During the training process, classification accuracy and AUROC were monitored. An early termination condition was set if AUROC did not improve over 20 epochs. After completing training, model performance was evaluated on a separate test dataset. Gradient-weighted class activation mapping (Grad-CAM) was utilized to identify important features and regions of interest influencing the decision of the classifier between input layer and the final convolution layer. 21 Then, heat maps of gradients were overlaid on the original image to visualize the spatial importance of predictions correlated with anatomical structures. For external validation, model performance was evaluated based on a prepared external test dataset collected from a different hospital to comply with the regulations of the Personal Information Protection Act. Statistical Analysis Statistical analyses were performed using the R software (version 4.0.2; https://www.r-project.org ). Patient age was presented as median (interquartile range, IQR) and analyzed using the Mann-Whitney U test because of the non-normal distribution confirmed by the Shapiro-Wilk test. Patient sex was presented as a number (percentage) and analyzed using the chi-square test. Case-control matching was performed using the MatchIt package on R. RESULTS The demographic information of the 166 patients enrolled in this study is summarized in Table 1 . The median age of both groups was identical at 60.5 years (range: 24–87 years). Of the 166 patients, 108 (65.1%) were female. None of the patients in the control group experienced facial pain. In aggregate, 108 and 58 patients experienced pain on the right and left sides, respectively. The distribution of pain in 58 patients involved any combination of two nerves (V1 and V2, V1 and V3, or V2 and V3). The V1 division was affected in four patients, V2 in 52, and V3 in 52. The duration of pain before the first outpatient visit averaged 4 years and ranged from less than 1 year to 30 years. 32 patients had experienced pain for more than 10 years. Seventy eight patients reported a Barrow Neurological Institute Pain Intensity (BNI) score of 3 at the time of outpatient visit, and their pain was controlled with medications. In contrast, 88 patients reported a BNI score of 4 or higher, and their pain was not adequately controlled even with medications. Of the 166 patients, 118 underwent microvascular decompression surgery for pain treatment, 18 underwent gamma knife surgery, and 30 were only prescribed medication. Table 1 Baseline clinical characteristics of study population (n = 166). Characteristics Trigeminal neuralgia (n = 166) Age, years Median (Range) 60.5 (24–87) Sex, n Male 58 (34.9%) Female 108 (65.1%) Direction Right 108 (65.1%) Left 58 (34.9%) Symptom Distribution V1 4 (2.4%) V2 52 (31.3%) V3 52 (31.3%) V1 + 2 20 (12.0%) V2 + 3 36 (21.7%) All branches 2 (1.2%) Symptom Duration * Under 10 year 134 (80.7%) Over 10 year 32 (19.3%) Median, year (Range) 4 (1 week − 30 year) Symptom Severity * BNI score 3 78 (47.0%) BNI score 4 60 (36.1%) BNI score 5 28 (16.9%) Treatment Medication 30 (18.1%) GKS 18 (10.8%) MVD 118 (71.1%) V1, ophthalmic branch; V2, maxillary branch; V3; mandibular branch; BNI, Barrow Neurological Institute; GKS, gamma knife surgery; MVD, microvascular decompression; *, from the first outpatient clinical date Among the different combinations of pretrained models, optimizers, and learning rates, DenseNet-121 with ImageNet initial weights optimized using Adam with an initial learning rate of 10 − 5 exhibited the best AUROC during our experiments. With this configuration, the training process was halted at the 66th epoch under the early termination condition, achieving a validation AUROC of 0.8095. The overall accuracy on the 133 images in the test dataset was 0.8722. The AUROC for predicting trigeminal neuralgia and control were 0.9006 and 0.8858, respectively. The trigeminal neuralgia prediction precision was 0.7500 and recall was 0.7273. The control prediction precision and recall were 0.9109 and 0.9200, respectively. The F1-scores for trigeminal neuralgia and control prediction were calculated to be 0.7385 and 0.9154, respectively. The weighted averages of the precision, recall and F1-score were 0.8710, 0.8722, and 0.8715, respectively. During external validation, the overall performance of the trained model exhibited a slight degradation. The AUROC for predicting trigeminal neuralgia was 0.8160, whereas that for predicting control was 0.8272. The precision and recall values for predicting trigeminal neuralgia were 0.6767 and 0.8800, respectively. The control prediction precision and recall were 0.8182 and 0.5400, respectively. F1-scores for trigeminal neuralgia and control were 0.7521 and 0.6506, respectively. The weighted averages of the precision, recall and F1-score were 0.7374, 0.7100, and 0.7014, respectively. The values of all performance indicators are summarized in Table 2 and Fig. 3 . During Grad CAM calculation, the principal attention area for predicting trigeminal neuralgia was mainly located around the sphenoid body and clivus in most TP cases, although heat maps exhibited more spacers in some cases (Fig. 4 A). Moreover, the sphenoid body accounted for a higher gradient distribution among the FP cases, with a wider attention area (Fig. 4 B). In contrast, control prediction was primarily based on the calvarium or cervical spine in both TN and FN cases. However, the trained model did not consider the sphenoid body in this case, which is an important factor in predicting trigeminal neuralgia (Figs. 4 C and D). Table 2 Model performance indicators Predicted Label TN Control Accuracy AUROC Precision Recall F1-score Institutional test set Actual label Trigeminal neuralgia ( 33 ) 24 9 0.8722 0.9006 0.7500 0.7273 0.7385 Control (100) 8 92 0.8858 0.9109 0.9200 0.7385 Weighted average 0.8710 0.8722 0.8715 External test set Actual label Trigeminal neuralgia (50) 44 6 0.7100 0.8160 0.6567 0.8800 0.7521 Control (50) 23 27 0.8272 0.8182 0.5400 0.6506 Weighted average 0.7374 0.7100 0.7014 AUROC, area under the receiver operating characteristic DISCUSSION Current diagnosis of trigeminal neuralgia relies heavily on the patient’s personal description of pain and symptoms. There are some established and specific diagnostic criteria for trigeminal neuralgia. 15,22,23 However, even with specific criteria, diagnosis can be challenging, especially in the early phases of the disorder. With over 60% of patients with trigeminal neuralgia experiencing misdiagnoses, there is a concerning trend of delays in obtaining accurate assessments. 24 Many patients undergo unnecessary treatments, such as dental procedures. 25 Conversely, in some cases, there is an overestimation of trigeminal neuralgia diagnoses, with only 21% eventually confirming the diagnosis according to international criteria. 26 This leads to the performance of potentially needless tests, such as magnetic resonance imaging (MRI) scans, compounding the complexity and cost of the diagnostic process. Despite recent attempts, such as the use of self-report questionnaires and artificial neural networks for trigeminal neuralgia screening, persistent challenges remain. Notably, even though an artificial neural network with sensitivity and specificity exceeding 80% in diagnosing episodic trigeminal neuralgia has been reported, 27 even this tool relies on the patient's description for diagnosis. In most scenarios, the diagnostic accuracy of trigeminal neuralgia falls short of expectations, highlighting the ongoing complexities associated with establishing a robust diagnostic approach. This underscores the need for objective diagnostic tools that emphasize the continuous pursuit of effective solutions in complex diagnostic situations. In light of the persistent challenges in diagnosing trigeminal neuralgia, our study aimed to explore alternative diagnostic methods. The application of deep learning to plain skull X-ray images, with a notable emphasis on the skull base, a region intricately linked to the trigeminal nerve and housing the Gasserian ganglion, yielded promising results. Interestingly, historical assertions dating back to the early 1900s suggested differences in the plane of the middle fossa and angle of the petrous process among individuals, 7 and recent advanced neuroimaging techniques have reaffirmed these distinctions, particularly in patients with trigeminal neuralgia. 13 Andrei et al. examined the role of the petrous bone, specifically the acute angle at the exit from Meckel’s cave, in trigeminal neuralgia pathogenesis. 28 Their findings underscored the implications of microvascular decompression procedures. A recent study explored sex-dependent anatomical variations in the posterior fossa, suggesting a potential link between the elevated prevalence of trigeminal neuralgia in females and distinct anatomical features. By integrating these insights, our study contributes to a comprehensive understanding of trigeminal neuralgia by bridging historical perspectives with contemporary anatomical nuances, yielding potential advancements in diagnosis and management. Despite well-established theories proclaiming NVC and thickened arachnoid membranes as contributors to trigeminal neuralgia 29–31 , explaining trigeminal neuralgia etiology solely based on variations in skull structure appears to be challenging. Nevertheless, our large sample study exploring deep learning yielded reliable results. To the best of our knowledge, this is the first study to have used deep learning to identify differences between the skulls of patients with trigeminal neuralgia and normal individuals. While our study suggests potential differences between two-dimensional skull images of patients with trigeminal neuralgia and normal individuals, further research is required to understand the effects of skull structural variations on the structural shaping of the brainstem and the course of the trigeminal nerve in a three-dimensional context. In fact, a study by Parise et al. analyzed the morphology of the posterior fossa using MRI, revealing that patients with trigeminal neuralgia tend to have smaller cerebellopontine angle cisterns than normal subjects. 13,32 This prompted additional investigations to determine whether the observed smaller cerebellopontine angle cisterns are indeed related to structural variations of the skull or if the latter influences the structure of the brain parenchyma, the morphology of the cerebrospinal fluid space, and/or surrounding vascular structures. Integration of deep learning into the diagnostic landscape has highlighted the potential alternative utility of plain skull X-ray imaging. A recent study at a neurosurgical center in Oslo highlighted a significant overestimation of trigeminal neuralgia diagnoses in patients referred for the first time. Only 21% of participants satisfied the confirmed diagnostic criteria, and over 60% of misdiagnosed patients exhibited an overestimated neurovascular contact on brain MRI, a key assessment factor for microvascular decompression. 24 Despite the utility of MRI in excluding differential diagnoses and identifying structural causes such as NVC, limitations regarding its cost and efficiency, particularly in terms of screening patients with facial pain for structural causes, are evident. Considering these factors, the use of artificial intelligence to screen patients with facial pain using plain skull X-ray images could revolutionize diagnostic approaches. Despite the significant promise of deep learning in plain skull X-ray imaging, it suffers from several limitations. The first limitation stems from the use of two-dimensional skull images, which does not allow accurate identification of specific regions and mechanisms contributing to the observed differences between patients with trigeminal neuralgia and normal individuals. Thorough understanding of these intricacies requires additional three-dimensional analyses, such as computed tomography or MRI. The second limitation concerns the choice of a control group in the deep learning model. In this study, patients with unruptured intracranial aneurysms were included in the control group, which introduced a potential bias in interpreting the findings. Ideally, the control group should consist of healthy individuals; however, obtaining their plain skull X-ray images is challenging. The selection of patients with unruptured intracranial aneurysms was influenced by the need to match sex ratio and age distribution with the trigeminal neuralgia cohort. Finally, given the retrospective study design and the exclusive enrollment of patients from the Asian population, the influence of racial differences on skull morphology and the results of this study require further investigation. 33 The focus on specific skull characteristics, such as dolichocephaly and brachycephaly, introduces the possibility of variations in the petrous bone angle with respect to the middle fossa plane. Further research is essential to address these potential influences and explore skull structural variations comprehensively in diverse populations. CONCLUSIONS The findings of this study highlight the promise of deep learning-based models in distinguishing between plain skull X-ray images of patients with trigeminal neuralgia and those belonging to a control group. Additional research is required to explore the impact of skull structural variations in three dimensions on brain parenchyma, course of the trigeminal nerve, and surrounding vascular structures. Declarations DATA AVAILABLITY Due to regulations imposed by hospitals and concerns regarding patient privacy, the raw datasets from individual clinical centers cannot be provided. The de-identified data is accessible for research purposes and can be obtained from the corresponding authors upon a reasonable request. Author Contribution Conception and design: H.T.C, Y.H.A., J.H.H., And S.Y.J., M.J.K. Funding acquisition: J.H.H., S.Y.J. Provision of data: J.E.K., J.B.P., H.K., K.H.H., C.Y.K., T.K.K., H.G.J. Data collection and cleaning: J.H.H., S.Y.J., J.E.K., J.B.P., H.K., K.H.H., C.Y.K., T.K.K., H.G.J. Data analysis and interpretation: M.J.K., J.B.P., T.K.K., J.H.H., S.Y.J., H.T.C.,Y.H.A. Manuscript writing and revising: all authors. Final approval of the manuscript: all authors. J.H.H., S.Y.J. and M.J.K. contributed equally as the first authors. H.T.C. and Y.H.A. contributed equally to this work and share corresponding authorship. ACKNOWLEDGEMENTS This study was supported by Grant No. 13-2023-0014 offered by Seoul National University Bundang Hospital Research Fund. Data Availability Due to regulations imposed by hospitals and concerns regarding patient privacy, the raw datasets from individual clinical centers cannot be provided. The de-identified data is accessible for research purposes and can be obtained from the corresponding authors upon a reasonable request. References Cruccu G, Di Stefano G, Truini A. Trigeminal Neuralgia. N Engl J Med 2020;383(8):754–762. (In eng). DOI: 10.1056/NEJMra1914484 . Maarbjerg S, Wolfram F, Gozalov A, Olesen J, Bendtsen L. Significance of neurovascular contact in classical trigeminal neuralgia. Brain 2015;138(Pt 2):311–9. (In eng). DOI: 10.1093/brain/awu349 . Lee A, McCartney S, Burbidge C, Raslan AM, Burchiel KJ. Trigeminal neuralgia occurs and recurs in the absence of neurovascular compression. J Neurosurg 2014;120(5):1048–54. (In eng). DOI: 10.3171/2014.1.Jns131410 . Katusic S, Beard CM, Bergstralh E, Kurland LT. Incidence and clinical features of trigeminal neuralgia, Rochester, Minnesota, 1945–1984. Ann Neurol 1990;27(1):89–95. (In eng). DOI: 10.1002/ana.410270114 . Reinard K, Nerenz DR, Basheer A, et al. Racial disparities in the diagnosis and management of trigeminal neuralgia. J Neurosurg 2017;126(2):368–374. (In eng). DOI: 10.3171/2015.11.Jns151177 . Siqueira SR, Teixeira MJ, Siqueira JT. Clinical characteristics of patients with trigeminal neuralgia referred to neurosurgery. Eur J Dent 2009;3(3):207–12. (In eng). Frazier CH. OPERATION FOR THE RADICAL CURE OF TRIGEMINAL NEURALGIA: ANALYSIS OF FIVE HUNDRED CASES. Ann Surg 1928;88(3):534–47. (In eng). DOI: 10.1097/00000658-192809000-00021 . KERR FWL. The Etiology of Trigeminal Neuralgia. Archives of Neurology 1963;8(1):15–25. DOI: 10.1001/archneur.1963.00460010031003 . Müller HR, Lévy A. Zur Pathogenese der Trigeminusneuralgie. Acta Neurochirurgica 1963;11(3):385–397. DOI: 10.1007/BF01404416 . Bjerrum J, Thornval G. Roentgenographic findings in trigeminal neuralgia. Acta radiol 1959;51(4):289–96. (In eng). DOI: 10.3109/00016925909171547 . Smith DG, Mumford JM. Petrous angle and trigeminal neuralgia. Pain 1980;8(3):269–277. (In eng). DOI: 10.1016/0304-3959(80)90072-x . Kundakçı YE, Doğan N, Kara İ, Öztürk M, Fazlıoğulları Z, Karabulut AK. Morphometric examination of trigeminal nerve and its adjacent structures in patients with trigeminal neuralgia: a case-control study. Turk J Med Sci 2022;52(5):1627–1638. (In eng). DOI: 10.55730/1300-0144.5503 . Parise M, Acioly MA, Ribeiro CT, Vincent M, Gasparetto EL. The role of the cerebellopontine angle cistern area and trigeminal nerve length in the pathogenesis of trigeminal neuralgia: a prospective case-control study. Acta Neurochir (Wien) 2013;155(5):863–8. (In eng). DOI: 10.1007/s00701-012-1573-0 . Kim T, Heo J, Jang DK, et al. Machine learning for detecting moyamoya disease in plain skull radiography using a convolutional neural network. EBioMedicine 2019;40:636–642. (In eng). DOI: 10.1016/j.ebiom.2018.12.043 . Cruccu G, Finnerup NB, Jensen TS, et al. Trigeminal neuralgia: New classification and diagnostic grading for practice and research. Neurology 2016;87(2):220–8. (In eng). DOI: 10.1212/WNL.0000000000002840 . Stuart EA, King G, Imai K, Ho D. MatchIt: nonparametric preprocessing for parametric causal inference. Journal of statistical software 2011. Huang G, Liu Z, Van Der Maaten L, Weinberger KQ. Densely connected convolutional networks. Proceedings of the IEEE conference on computer vision and pattern recognition2017:4700–4708. Russakovsky O, Deng J, Su H, et al. Imagenet large scale visual recognition challenge. International journal of computer vision 2015;115(3):211–252. Rajpurkar P, Irvin J, Zhu K, et al. Chexnet: Radiologist-level pneumonia detection on chest x-rays with deep learning. arXiv preprint arXiv:171105225 2017. Pizer SM, Amburn EP, Austin JD, et al. Adaptive histogram equalization and its variations. Computer vision, graphics, and image processing 1987;39(3):355–368. Selvaraju RR, Das A, Vedantam R, Cogswell M, Parikh D, Batra D. Grad-CAM: Why did you say that? arXiv preprint arXiv:161107450 2016. The International Classification of Headache Disorders, 3rd edition (beta version). Cephalalgia 2013;33(9):629–808. (In eng). DOI: 10.1177/0333102413485658 . Aziz Q, Giamberardino MA, Barke A, et al. The IASP classification of chronic pain for ICD-11: chronic secondary visceral pain. Pain 2019;160(1):69–76. (In eng). DOI: 10.1097/j.pain.0000000000001362 . Cheraghi-Sohi S, Holland F, Reeves D, et al. The incidence of diagnostic errors in UK primary care and implications for health care, research, and medical education: a retrospective record analysis of missed diagnostic opportunities. British Journal of General Practice 2018;68:bjgp18X696857. DOI: 10.3399/bjgp18X696857 . von Eckardstein KL, Keil M, Rohde V. Unnecessary dental procedures as a consequence of trigeminal neuralgia. Neurosurg Rev 2015;38(2):355–60; discussion 360. (In eng). DOI: 10.1007/s10143-014-0591-1 . Slettebø H. Is this really trigeminal neuralgia? Diagnostic re-evaluation of patients referred for neurosurgery. Scand J Pain 2021;21(4):788–793. (In eng). DOI: 10.1515/sjpain-2021-0045 . McCartney S, Weltin M, Burchiel KJ. Use of an artificial neural network for diagnosis of facial pain syndromes: an update. Stereotact Funct Neurosurg 2014;92(1):44–52. (In eng). DOI: 10.1159/000353188 . Brinzeu A, Dumot C, Sindou M. Role of the petrous ridge and angulation of the trigeminal nerve in the pathogenesis of trigeminal neuralgia, with implications for microvascular decompression. Acta Neurochir (Wien) 2018;160(5):971–976. (In eng). DOI: 10.1007/s00701-018-3468-1 . Jannetta PJ. Vascular compression is the cause of trigeminal neuralgia. APS Journal 1993;2(4):217–227. DOI: https://doi.org/10.1016/S1058-9139(05)80246-3 . Sindou M, Howeidy T, Acevedo G. Anatomical observations during microvascular decompression for idiopathic trigeminal neuralgia (with correlations between topography of pain and site of the neurovascular conflict). Prospective study in a series of 579 patients. Acta Neurochir (Wien) 2002;144(1):1–12; discussion 12 – 3. (In eng). DOI: 10.1007/s701-002-8269-4 . Ishikawa M, Nishi S, Aoki T, et al. Operative findings in cases of trigeminal neuralgia without vascular compression: proposal of a different mechanism. J Clin Neurosci 2002;9(2):200–4. (In eng). DOI: 10.1054/jocn.2001.0922 . Parise M, Ribeiro CT, Vincent M, Gasparetto E. Small cerebellopontine angle cisterns inpatients with trigeminal neuralgia. J Neurosurg 2010;112(4):900–1; author reply 901. (In eng). DOI: 10.3171/2009.10.Jns091517 . Kuroe K, Rosas A, Molleson T. Variation in the cranial base orientation and facial skeleton in dry skulls sampled from three major populations. Eur J Orthod 2004;26(2):201–7. (In eng). DOI: 10.1093/ejo/26.2.201 . Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 29 May, 2025 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 14 Oct, 2024 Reviews received at journal 10 Oct, 2024 Reviews received at journal 23 Sep, 2024 Reviewers agreed at journal 18 Sep, 2024 Reviewers agreed at journal 18 Sep, 2024 Reviewers invited by journal 16 Sep, 2024 Editor assigned by journal 16 Sep, 2024 Editor invited by journal 28 Jul, 2024 Submission checks completed at journal 25 Jul, 2024 First submitted to journal 20 Jul, 2024 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4775021","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":342809783,"identity":"6a6fe862-74d6-442a-b657-8d83ef45427a","order_by":0,"name":"Jung Ho Han","email":"","orcid":"","institution":"Department of neurosurgery, Seoul National University Bundang Hospital","correspondingAuthor":false,"prefix":"","firstName":"Jung","middleName":"Ho","lastName":"Han","suffix":""},{"id":342809784,"identity":"f2e09663-6a81-4eaf-8366-b7ee1db88ca4","order_by":1,"name":"So Young Ji","email":"","orcid":"","institution":"Department of neurosurgery, Seoul National University Bundang Hospital","correspondingAuthor":false,"prefix":"","firstName":"So","middleName":"Young","lastName":"Ji","suffix":""},{"id":342809785,"identity":"dafd31ab-ea41-4fff-86f4-b6bf6ce90be5","order_by":2,"name":"Myung Ju Kim","email":"","orcid":"","institution":"Center for Artificial Intelligence in Healthcare, Seoul National University Bundang Hospital","correspondingAuthor":false,"prefix":"","firstName":"Myung","middleName":"Ju","lastName":"Kim","suffix":""},{"id":342809786,"identity":"6a99e48e-ec50-4d1b-94e6-ea2f59098b3d","order_by":3,"name":"Ji Eyon Kwon","email":"","orcid":"","institution":"Department of neurosurgery, Seoul National University Bundang Hospital","correspondingAuthor":false,"prefix":"","firstName":"Ji","middleName":"Eyon","lastName":"Kwon","suffix":""},{"id":342809787,"identity":"c8b95b93-7ee4-4bbb-a0b0-6f0e40aae210","order_by":4,"name":"Jin Byung Park","email":"","orcid":"","institution":"Center for Artificial Intelligence in Healthcare, Seoul National University Bundang Hospital","correspondingAuthor":false,"prefix":"","firstName":"Jin","middleName":"Byung","lastName":"Park","suffix":""},{"id":342809788,"identity":"e6036a20-927a-453f-b810-48a3888aeb34","order_by":5,"name":"Ho Kang","email":"","orcid":"","institution":"Department of neurosurgery, Seoul National University Bundang Hospital","correspondingAuthor":false,"prefix":"","firstName":"Ho","middleName":"","lastName":"Kang","suffix":""},{"id":342809790,"identity":"5e704f36-15c0-442e-940d-50d375c3cc39","order_by":6,"name":"Kihwan Hwang","email":"","orcid":"","institution":"Department of neurosurgery, Seoul National University Bundang Hospital","correspondingAuthor":false,"prefix":"","firstName":"Kihwan","middleName":"","lastName":"Hwang","suffix":""},{"id":342809791,"identity":"f5570828-5a5c-4d27-911a-a64f93e2d9e8","order_by":7,"name":"Chae-Yong Kim","email":"","orcid":"","institution":"Department of neurosurgery, Seoul National University Bundang Hospital","correspondingAuthor":false,"prefix":"","firstName":"Chae-Yong","middleName":"","lastName":"Kim","suffix":""},{"id":342809792,"identity":"71b84fd3-c604-43dd-8f8b-2dab3f6f891f","order_by":8,"name":"Tackeun Kim","email":"","orcid":"","institution":"TALOS Corp.","correspondingAuthor":false,"prefix":"","firstName":"Tackeun","middleName":"","lastName":"Kim","suffix":""},{"id":342809793,"identity":"9d50b412-4c60-4d02-a28e-56484283ba1b","order_by":9,"name":"Han-Gil Jeong","email":"","orcid":"","institution":"Department of neurosurgery, Seoul National University Bundang Hospital","correspondingAuthor":false,"prefix":"","firstName":"Han-Gil","middleName":"","lastName":"Jeong","suffix":""},{"id":342809795,"identity":"bc51b8c0-2c24-4077-8aeb-91fd8c0ac8b6","order_by":10,"name":"Young Hwan Ahn","email":"","orcid":"","institution":"Department of Neurosurgery, Ajou University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Young","middleName":"Hwan","lastName":"Ahn","suffix":""},{"id":342809796,"identity":"16f5f9fb-bdaa-439d-b76c-bf491bec6cc2","order_by":11,"name":"Hyun-Tai Chung","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAwElEQVRIiWNgGAWjYBACAwYGNhAthyRASAsbRIsx6VoSG4jWYi7f/Owx747a9PmzewwYftQwGJs3ENBi2cZmbsx75njuhjtnDBh7jjGYyRwg5LBjPGzSvG3HcjdI5Bgw8DYw2EgQchhMS7r8jBwDxr8kaKlJYLiRY8AMtMWMCC1pZpJz2w4YbriRVnBY5piEMWEthw8/k3jbVicvPyN548M3NTaGMwhpgYLDYPIAAwNBO+CgjmiVo2AUjIJRMAIBAOVQN26R+SbIAAAAAElFTkSuQmCC","orcid":"","institution":"Department of Neurosurgery, Seoul National University Hospital","correspondingAuthor":true,"prefix":"","firstName":"Hyun-Tai","middleName":"","lastName":"Chung","suffix":""}],"badges":[],"createdAt":"2024-07-21 02:44:16","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4775021/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4775021/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-025-03254-7","type":"published","date":"2025-05-29T15:57:52+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":63422491,"identity":"a9bd9877-e4fb-4869-a967-209e5948f675","added_by":"auto","created_at":"2024-08-28 02:52:01","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":2759017,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart shows the process of enrollment of patients and partitioning.\u003c/p\u003e","description":"","filename":"Figure1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4775021/v1/cbf4b2a5d12a9fa68296357e.jpg"},{"id":63422490,"identity":"73a87af4-81ff-46d0-9e98-badb12afa217","added_by":"auto","created_at":"2024-08-28 02:52:01","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":736204,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart shows each augmentation step and related example images. The images are slightly blurred with gausian filtering to enhance anonymity. In the actual investigation, blurring was not applied to the images. CLAHE, contrast limited adaptive histogram equalization.\u003c/p\u003e","description":"","filename":"Figure2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4775021/v1/4f8ef44c25d9e94ae335e730.jpg"},{"id":63422492,"identity":"b1007d13-1e0d-4bb5-b241-017800f15583","added_by":"auto","created_at":"2024-08-28 02:52:01","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":3337677,"visible":true,"origin":"","legend":"\u003cp\u003ePlots show area under the receiver operating characteristic of trained model to classify test set (A) and external test set (B). Class 1 and 0 refer trigeminal neuralgia and control, respectively. ROC, receiver operating characteristic\u003c/p\u003e","description":"","filename":"Figure3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4775021/v1/16371a26f47981eae8f06fa8.jpg"},{"id":63422493,"identity":"c24bcadf-9077-41a9-a9fb-cadbe18113d0","added_by":"auto","created_at":"2024-08-28 02:52:01","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":2731220,"visible":true,"origin":"","legend":"\u003cp\u003eImage matrix shows the heatmaps for decision of trained model by Grad-CAM application. In the case of a prediction as trigeminal neuralgia, higher gradient heatmaps are concentrated around the sphenoid body to clivus. However, false positive cases show that the area is more spreadout (A\u0026amp;B). Heatmaps for predicting control is concentrated at the calvarium or the cervical spines (C\u0026amp;D). The skull images are slightly blurred with gausian filtering to enhance anonymity. In the actual investigation, blurring was not applied to the images.\u003c/p\u003e","description":"","filename":"Figure4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4775021/v1/c5ec817293892df11660a5be.jpg"},{"id":83782969,"identity":"0d017769-f5b7-4652-ab4a-0d79f61b1cb7","added_by":"auto","created_at":"2025-06-02 16:09:26","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":10122486,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4775021/v1/da19826f-904b-468b-af25-f5d9dd9a47ac.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Diagnosis of trigeminal neuralgia based on plain skull radiography using convolutional neural network","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003eTrigeminal neuralgia is characterized by recurrent, unilateral, short bursts of electric shock-like pain affecting one or more branches of the trigeminal nerve.\u003csup\u003e1\u003c/sup\u003e The most widely accepted theory regarding the cause of trigeminal neuralgia is neurovascular compression (NVC).\u003csup\u003e2\u003c/sup\u003e However, recent studies have challenged this notion, suggesting that NVC may not be a sufficient or necessary condition for trigeminal neuralgia development,\u003csup\u003e3\u003c/sup\u003e prompting a closer examination of additional predisposing factors. Higher prevalence of trigeminal neuralgia in females and Caucasian individuals further points to the potential involvement of other etiological factors.\u003csup\u003e4\u0026ndash;6\u003c/sup\u003e\u003c/p\u003e \u003cp\u003eBefore the acceptance of NVC as a cause of trigeminal neuralgia during the 1960s and 70s, structural changes in the skull were considered mechanical factors in trigeminal neuralgia etiology.\u003csup\u003e7,8\u003c/sup\u003e Early studies using plain skull X-ray imaging identified morphological differences associated with trigeminal neuralgia, such as bony angles at the ridge of the petrous bone.\u003csup\u003e9,10,11\u003c/sup\u003e With advances in imaging technology, recent studies have explored the role of neuroanatomical structures in trigeminal neuralgia etiology, revealing smaller posterior fossa and cistern volumes and increased crowding in patients with trigeminal neuralgia.\u003csup\u003e12,13\u003c/sup\u003e These findings suggest that there are additional structural features that distinguish patients with trigeminal neuralgia from normal individuals.\u003c/p\u003e \u003cp\u003eAlthough the prevailing consensus on trigeminal neuralgia etiology leans towards NVC, the practical utility of discerning subtle differences in plain skull X-ray images with the naked eye has diminished over time. However, with recent breakthroughs in image analysis using deep learning, renewed exploration of these claims is warranted. This technology enables novel re-evaluation of the significance of structural variations in trigeminal neuralgia, potentially providing valuable insights that were previously too challenging to ascertain. In the realm of medical imaging, deep learning, particularly convolutional neural networks (CNN), has exhibited great promise in recognizing features that may elude human observation.\u003csup\u003e14\u003c/sup\u003e Therefore, in this study, we evaluated the applicability of trigeminal neuralgia in distinguishing plain skull X-ray images using CNN.\u003c/p\u003e"},{"header":"MATERIALS AND METHODS","content":"\u003cp\u003eStudy enrollment\u003c/p\u003e \u003cp\u003e The study design was reviewed and approved by the Institutional Review Boards of Seoul National University Bundang Hospital (B-1910-572-101) and Ajou University Medical Center (AJIRB-MED-MDB-22-234), and it adhered to the recommendations of the Declaration of Helsinki for biomedical research involving human subjects. This study complied with all IRB ethical regulations. Informed consents were waived by the Committee due to the retrospective and deidentified nature of the data. During the collection of lateral skull radiographs for investigation, the following inclusion criteria were used: 1) patients aged over 16 years, and 2) patients with lateral skull radiographs without any trace of previous craniofacial surgeries or any truncation of the image. We enrolled clinically established patients with trigeminal neuralgia experiencing paroxysmal unilateral orofacial pain distributed along the trigeminal territory which should be triggered by typical maneuvers.\u003csup\u003e15\u003c/sup\u003e To identify eligible patients with trigeminal neuralgia, we used trigeminal neuralgia registry and medical records. In aggregate, 277 patients were diagnosed with trigeminal neuralgia between January 2013 and June 2020. As a control group, we selected patients with unruptured intracranial aneurysms who underwent clipping surgery because most patients underwent skull radiography for preoperative planning. In aggregate, 3,045 patients were identified using a clinical data warehouse query system. Among them, 2,706 underwent skull radiography and were ready for review. After reviewing all medical records and skull radiographs, 166 patients with trigeminal neuralgia and 1,702 control group patients were retained as qualifying study subjects with appropriate skull X-ray images. To control confounding effects, we performed three-fold matching by age and sex.\u003csup\u003e16\u003c/sup\u003e As a result, 664 patients (166 trigeminal neuralgias and 498 controls) were enrolled in this study. All subjects in the control group were confirmed to not have undergone Rany procedures or surgeries related to facial pain and exhibited no history of related prescriptions.\u003c/p\u003e \u003cp\u003eEach group was partitioned into training, validation, and test datasets in a 6:2:2 ratio using random permutation. Generalized performance of the CNN was evaluated on an external test dataset consisting of 100 patients (50 trigeminal neuralgias and 50 controls) from other institutes, in the same manner as described above. The patient selection process is illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eConvolutional neural network classifier\u003c/p\u003e \u003cp\u003eWe used TensorFlow (version 2.0.0, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.tensorflow.org/\u003c/span\u003e\u003cspan address=\"https://www.tensorflow.org/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) as a framework to train and evaluate the neural network model in skull radiograph classification. Pre-trained \u0026lsquo;Densely Connected Convolutional Network\u0026rsquo; (DenseNet-121)\u003csup\u003e17\u003c/sup\u003e with ImageNet dataset\u003csup\u003e18\u003c/sup\u003e and ChesXNet dataset\u003csup\u003e19\u003c/sup\u003e were imported for initial weight configuration. Originally, the pre-trained DenseNet-121 was designed to identify 1,000 (ImageNet) and 14 (ChesXNet) labels using 224 \u0026times; 224 images. However, as we required a binominal classifier, the top layers, consisting of 1,000 and 14 fully connected nodes, respectively, were replaced with layers that included two nodes. The original softmax activation function was replaced by the sigmoid function \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:(\\:f\\left(x\\right)=\\:\\frac{1}{1+{e}^{-x}}\\:)\\)\u003c/span\u003e\u003c/span\u003e, with 7,039,554 parameters. The loss function was set to be a binary cross-entropy ( \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:-{y}_{true}\\text{log}f\\left(x\\right)\\:-\\left(1-{y}_{true}\\right)\\text{log}(1-f\\left(x\\right)\\:)\\)\u003c/span\u003e\u003c/span\u003e ). Various optimizers (SGD, RMSprop, Adagrad, Adadelta, Adam, Adamax, and Nadam), with initial learning rates ranging between 10\u003csup\u003e\u0026minus;6\u003c/sup\u003e and 10\u003csup\u003e\u0026minus;3\u003c/sup\u003e, were tested to achieve best model performance.\u003c/p\u003e \u003cp\u003ePreprocessing, model training, and evaluation\u003c/p\u003e \u003cp\u003eConsidering the modest number of data points in the training dataset, image augmentation was performed on it in two ways. First, contrast-limited adaptive histogram equalization was applied to enhance particular characteristics by improving image contrast.\u003csup\u003e20\u003c/sup\u003e Second, rotation within 15\u0026deg;, translation within 10%, transposition within 10%, or brightness alteration within 20% were randomly applied in combination (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). Moreover, to account for class imbalance (1:3), the control group was augmented by a factor of 14 to 4,172 images and the trigeminal neuralgia group by a factor of 20 to 2,000 images. For the validation dataset, only contrast-limited adaptive histogram equalization was applied to produce two iterations for each label. The test dataset was used without any modifications.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eSeveral performance indicators were calculated from the confusion matrix, which consisted of the numbers of true positives (TP, predicts TN as TN), true negatives (TN, predicts control as control), false positives (FP, predicts control as TN), and false negatives (FN, predicts TN as control). The accuracy, precision, and recall were defined to be \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\frac{TP+TN}{TP+FP+TN+FN}\\)\u003c/span\u003e\u003c/span\u003e, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\frac{TP}{TP+FP}\\)\u003c/span\u003e\u003c/span\u003e, and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\frac{TP}{TP+FN}\\)\u003c/span\u003e\u003c/span\u003e, respectively. The F1-score was defined as the harmonic mean of precision and recall. The area under the receiver-operating characteristic curve (AUROC) was calculated using the probability produced by the sigmoid output for each label.\u003c/p\u003e \u003cp\u003eDuring the training process, classification accuracy and AUROC were monitored. An early termination condition was set if AUROC did not improve over 20 epochs. After completing training, model performance was evaluated on a separate test dataset. Gradient-weighted class activation mapping (Grad-CAM) was utilized to identify important features and regions of interest influencing the decision of the classifier between input layer and the final convolution layer.\u003csup\u003e21\u003c/sup\u003e Then, heat maps of gradients were overlaid on the original image to visualize the spatial importance of predictions correlated with anatomical structures.\u003c/p\u003e \u003cp\u003eFor external validation, model performance was evaluated based on a prepared external test dataset collected from a different hospital to comply with the regulations of the Personal Information Protection Act.\u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eStatistical Analysis\u003c/h2\u003e \u003cp\u003eStatistical analyses were performed using the R software (version 4.0.2; \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.r-project.org\u003c/span\u003e\u003cspan address=\"https://www.r-project.org\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). Patient age was presented as median (interquartile range, IQR) and analyzed using the Mann-Whitney U test because of the non-normal distribution confirmed by the Shapiro-Wilk test. Patient sex was presented as a number (percentage) and analyzed using the chi-square test. Case-control matching was performed using the MatchIt package on R.\u003c/p\u003e \u003c/div\u003e"},{"header":"RESULTS","content":"\u003cp\u003eThe demographic information of the 166 patients enrolled in this study is summarized in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. The median age of both groups was identical at 60.5 years (range: 24\u0026ndash;87 years). Of the 166 patients, 108 (65.1%) were female. None of the patients in the control group experienced facial pain. In aggregate, 108 and 58 patients experienced pain on the right and left sides, respectively. The distribution of pain in 58 patients involved any combination of two nerves (V1 and V2, V1 and V3, or V2 and V3). The V1 division was affected in four patients, V2 in 52, and V3 in 52. The duration of pain before the first outpatient visit averaged 4 years and ranged from less than 1 year to 30 years. 32 patients had experienced pain for more than 10 years. Seventy eight patients reported a Barrow Neurological Institute Pain Intensity (BNI) score of 3 at the time of outpatient visit, and their pain was controlled with medications. In contrast, 88 patients reported a BNI score of 4 or higher, and their pain was not adequately controlled even with medications. Of the 166 patients, 118 underwent microvascular decompression surgery for pain treatment, 18 underwent gamma knife surgery, and 30 were only prescribed medication.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eBaseline clinical characteristics of study population (n\u0026thinsp;=\u0026thinsp;166).\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCharacteristics\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTrigeminal neuralgia\u003c/p\u003e \u003cp\u003e (n\u0026thinsp;=\u0026thinsp;166)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge, years\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMedian (Range)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e60.5 (24\u0026ndash;87)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSex, n\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e58 (34.9%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e108 (65.1%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDirection\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRight\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e108 (65.1%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLeft\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e58 (34.9%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSymptom Distribution\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eV1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4 (2.4%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eV2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e52 (31.3%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eV3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e52 (31.3%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eV1\u0026thinsp;+\u0026thinsp;2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e20 (12.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eV2\u0026thinsp;+\u0026thinsp;3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e36 (21.7%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAll branches\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2 (1.2%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSymptom Duration *\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUnder 10 year\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e134 (80.7%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOver 10 year\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e32 (19.3%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMedian, year (Range)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4 (1 week \u0026minus;\u0026thinsp;30 year)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSymptom Severity *\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBNI score 3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e78 (47.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBNI score 4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e60 (36.1%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBNI score 5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e28 (16.9%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTreatment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMedication\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e30 (18.1%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGKS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e18 (10.8%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMVD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e118 (71.1%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"2\"\u003eV1, ophthalmic branch; V2, maxillary branch; V3; mandibular branch; BNI, Barrow Neurological Institute; GKS, gamma knife surgery; MVD, microvascular decompression; *, from the first outpatient clinical date\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eAmong the different combinations of pretrained models, optimizers, and learning rates, DenseNet-121 with ImageNet initial weights optimized using Adam with an initial learning rate of 10\u003csup\u003e\u0026minus;\u0026thinsp;5\u003c/sup\u003e exhibited the best AUROC during our experiments. With this configuration, the training process was halted at the 66th epoch under the early termination condition, achieving a validation AUROC of 0.8095.\u003c/p\u003e \u003cp\u003eThe overall accuracy on the 133 images in the test dataset was 0.8722. The AUROC for predicting trigeminal neuralgia and control were 0.9006 and 0.8858, respectively. The trigeminal neuralgia prediction precision was 0.7500 and recall was 0.7273. The control prediction precision and recall were 0.9109 and 0.9200, respectively. The F1-scores for trigeminal neuralgia and control prediction were calculated to be 0.7385 and 0.9154, respectively. The weighted averages of the precision, recall and F1-score were 0.8710, 0.8722, and 0.8715, respectively.\u003c/p\u003e \u003cp\u003eDuring external validation, the overall performance of the trained model exhibited a slight degradation. The AUROC for predicting trigeminal neuralgia was 0.8160, whereas that for predicting control was 0.8272. The precision and recall values for predicting trigeminal neuralgia were 0.6767 and 0.8800, respectively. The control prediction precision and recall were 0.8182 and 0.5400, respectively. F1-scores for trigeminal neuralgia and control were 0.7521 and 0.6506, respectively. The weighted averages of the precision, recall and F1-score were 0.7374, 0.7100, and 0.7014, respectively. The values of all performance indicators are summarized in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e and Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. During Grad CAM calculation, the principal attention area for predicting trigeminal neuralgia was mainly located around the sphenoid body and clivus in most TP cases, although heat maps exhibited more spacers in some cases (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eA). Moreover, the sphenoid body accounted for a higher gradient distribution among the FP cases, with a wider attention area (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eB). In contrast, control prediction was primarily based on the calvarium or cervical spine in both TN and FN cases. However, the trained model did not consider the sphenoid body in this case, which is an important factor in predicting trigeminal neuralgia (Figs.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eC and D).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eModel performance indicators\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"11\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c11\" colnum=\"11\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003ePredicted Label\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c10\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c11\"\u003e\u0026nbsp;\u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eControl\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eAUROC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003ePrecision\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003eRecall\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003eF1-score\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eInstitutional test set\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eActual label\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eTrigeminal neuralgia (\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e24\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.8722\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.9006\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.7500\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.7273\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.7385\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eControl (100)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.8858\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.9109\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.9200\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.7385\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c7\" namest=\"c6\"\u003e \u003cp\u003eWeighted average\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.8710\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.8722\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.8715\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eExternal test set\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" morerows=\"1\" nameend=\"c2\" namest=\"c1\" rowspan=\"2\"\u003e \u003cp\u003eActual label\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTrigeminal neuralgia (50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e44\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.7100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.8160\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.6567\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.8800\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.7521\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eControl (50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e23\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e27\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.8272\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.8182\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.5400\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.6506\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c7\" namest=\"c6\"\u003e \u003cp\u003eWeighted average\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.7374\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.7100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.7014\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"11\"\u003eAUROC, area under the receiver operating characteristic\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003eCurrent diagnosis of trigeminal neuralgia relies heavily on the patient\u0026rsquo;s personal description of pain and symptoms. There are some established and specific diagnostic criteria for trigeminal neuralgia.\u003csup\u003e15,22,23\u003c/sup\u003e However, even with specific criteria, diagnosis can be challenging, especially in the early phases of the disorder. With over 60% of patients with trigeminal neuralgia experiencing misdiagnoses, there is a concerning trend of delays in obtaining accurate assessments.\u003csup\u003e24\u003c/sup\u003e Many patients undergo unnecessary treatments, such as dental procedures.\u003csup\u003e25\u003c/sup\u003e Conversely, in some cases, there is an overestimation of trigeminal neuralgia diagnoses, with only 21% eventually confirming the diagnosis according to international criteria.\u003csup\u003e26\u003c/sup\u003e This leads to the performance of potentially needless tests, such as magnetic resonance imaging (MRI) scans, compounding the complexity and cost of the diagnostic process.\u003c/p\u003e \u003cp\u003eDespite recent attempts, such as the use of self-report questionnaires and artificial neural networks for trigeminal neuralgia screening, persistent challenges remain. Notably, even though an artificial neural network with sensitivity and specificity exceeding 80% in diagnosing episodic trigeminal neuralgia has been reported,\u003csup\u003e27\u003c/sup\u003e even this tool relies on the patient's description for diagnosis. In most scenarios, the diagnostic accuracy of trigeminal neuralgia falls short of expectations, highlighting the ongoing complexities associated with establishing a robust diagnostic approach. This underscores the need for objective diagnostic tools that emphasize the continuous pursuit of effective solutions in complex diagnostic situations.\u003c/p\u003e \u003cp\u003eIn light of the persistent challenges in diagnosing trigeminal neuralgia, our study aimed to explore alternative diagnostic methods. The application of deep learning to plain skull X-ray images, with a notable emphasis on the skull base, a region intricately linked to the trigeminal nerve and housing the Gasserian ganglion, yielded promising results. Interestingly, historical assertions dating back to the early 1900s suggested differences in the plane of the middle fossa and angle of the petrous process among individuals,\u003csup\u003e7\u003c/sup\u003e and recent advanced neuroimaging techniques have reaffirmed these distinctions, particularly in patients with trigeminal neuralgia.\u003csup\u003e13\u003c/sup\u003e\u003c/p\u003e \u003cp\u003eAndrei et al. examined the role of the petrous bone, specifically the acute angle at the exit from Meckel\u0026rsquo;s cave, in trigeminal neuralgia pathogenesis.\u003csup\u003e28\u003c/sup\u003e Their findings underscored the implications of microvascular decompression procedures. A recent study explored sex-dependent anatomical variations in the posterior fossa, suggesting a potential link between the elevated prevalence of trigeminal neuralgia in females and distinct anatomical features. By integrating these insights, our study contributes to a comprehensive understanding of trigeminal neuralgia by bridging historical perspectives with contemporary anatomical nuances, yielding potential advancements in diagnosis and management.\u003c/p\u003e \u003cp\u003eDespite well-established theories proclaiming NVC and thickened arachnoid membranes as contributors to trigeminal neuralgia\u003csup\u003e29\u0026ndash;31\u003c/sup\u003e, explaining trigeminal neuralgia etiology solely based on variations in skull structure appears to be challenging. Nevertheless, our large sample study exploring deep learning yielded reliable results. To the best of our knowledge, this is the first study to have used deep learning to identify differences between the skulls of patients with trigeminal neuralgia and normal individuals. While our study suggests potential differences between two-dimensional skull images of patients with trigeminal neuralgia and normal individuals, further research is required to understand the effects of skull structural variations on the structural shaping of the brainstem and the course of the trigeminal nerve in a three-dimensional context. In fact, a study by Parise et al. analyzed the morphology of the posterior fossa using MRI, revealing that patients with trigeminal neuralgia tend to have smaller cerebellopontine angle cisterns than normal subjects.\u003csup\u003e13,32\u003c/sup\u003e This prompted additional investigations to determine whether the observed smaller cerebellopontine angle cisterns are indeed related to structural variations of the skull or if the latter influences the structure of the brain parenchyma, the morphology of the cerebrospinal fluid space, and/or surrounding vascular structures.\u003c/p\u003e \u003cp\u003eIntegration of deep learning into the diagnostic landscape has highlighted the potential alternative utility of plain skull X-ray imaging. A recent study at a neurosurgical center in Oslo highlighted a significant overestimation of trigeminal neuralgia diagnoses in patients referred for the first time. Only 21% of participants satisfied the confirmed diagnostic criteria, and over 60% of misdiagnosed patients exhibited an overestimated neurovascular contact on brain MRI, a key assessment factor for microvascular decompression.\u003csup\u003e24\u003c/sup\u003e Despite the utility of MRI in excluding differential diagnoses and identifying structural causes such as NVC, limitations regarding its cost and efficiency, particularly in terms of screening patients with facial pain for structural causes, are evident. Considering these factors, the use of artificial intelligence to screen patients with facial pain using plain skull X-ray images could revolutionize diagnostic approaches.\u003c/p\u003e \u003cp\u003eDespite the significant promise of deep learning in plain skull X-ray imaging, it suffers from several limitations. The first limitation stems from the use of two-dimensional skull images, which does not allow accurate identification of specific regions and mechanisms contributing to the observed differences between patients with trigeminal neuralgia and normal individuals. Thorough understanding of these intricacies requires additional three-dimensional analyses, such as computed tomography or MRI. The second limitation concerns the choice of a control group in the deep learning model. In this study, patients with unruptured intracranial aneurysms were included in the control group, which introduced a potential bias in interpreting the findings. Ideally, the control group should consist of healthy individuals; however, obtaining their plain skull X-ray images is challenging. The selection of patients with unruptured intracranial aneurysms was influenced by the need to match sex ratio and age distribution with the trigeminal neuralgia cohort. Finally, given the retrospective study design and the exclusive enrollment of patients from the Asian population, the influence of racial differences on skull morphology and the results of this study require further investigation.\u003csup\u003e33\u003c/sup\u003e The focus on specific skull characteristics, such as dolichocephaly and brachycephaly, introduces the possibility of variations in the petrous bone angle with respect to the middle fossa plane. Further research is essential to address these potential influences and explore skull structural variations comprehensively in diverse populations.\u003c/p\u003e"},{"header":"CONCLUSIONS","content":"\u003cp\u003eThe findings of this study highlight the promise of deep learning-based models in distinguishing between plain skull X-ray images of patients with trigeminal neuralgia and those belonging to a control group. Additional research is required to explore the impact of skull structural variations in three dimensions on brain parenchyma, course of the trigeminal nerve, and surrounding vascular structures.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eDATA AVAILABLITY\u003c/h2\u003e \u003cp\u003eDue to regulations imposed by hospitals and concerns regarding patient privacy, the raw datasets from individual clinical centers cannot be provided. The de-identified data is accessible for research purposes and can be obtained from the corresponding authors upon a reasonable request.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eConception and design: H.T.C, Y.H.A., J.H.H., And S.Y.J., M.J.K. Funding acquisition: J.H.H., S.Y.J. Provision of data: J.E.K., J.B.P., H.K., K.H.H., C.Y.K., T.K.K., H.G.J. Data collection and cleaning: J.H.H., S.Y.J., J.E.K., J.B.P., H.K., K.H.H., C.Y.K., T.K.K., H.G.J. Data analysis and interpretation: M.J.K., J.B.P., T.K.K., J.H.H., S.Y.J., H.T.C.,Y.H.A. Manuscript writing and revising: all authors. Final approval of the manuscript: all authors. J.H.H., S.Y.J. and M.J.K. contributed equally as the first authors. H.T.C. and Y.H.A. contributed equally to this work and share corresponding authorship.\u003c/p\u003e\u003ch2\u003eACKNOWLEDGEMENTS\u003c/h2\u003e \u003cp\u003eThis study was supported by Grant No. 13-2023-0014 offered by Seoul National University Bundang Hospital Research Fund.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eDue to regulations imposed by hospitals and concerns regarding patient privacy, the raw datasets from individual clinical centers cannot be provided. The de-identified data is accessible for research purposes and can be obtained from the corresponding authors upon a reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eCruccu G, Di Stefano G, Truini A. Trigeminal Neuralgia. N Engl J Med 2020;383(8):754\u0026ndash;762. (In eng). DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1056/NEJMra1914484\u003c/span\u003e\u003cspan address=\"10.1056/NEJMra1914484\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMaarbjerg S, Wolfram F, Gozalov A, Olesen J, Bendtsen L. Significance of neurovascular contact in classical trigeminal neuralgia. Brain 2015;138(Pt 2):311\u0026ndash;9. (In eng). DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/brain/awu349\u003c/span\u003e\u003cspan address=\"10.1093/brain/awu349\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee A, McCartney S, Burbidge C, Raslan AM, Burchiel KJ. Trigeminal neuralgia occurs and recurs in the absence of neurovascular compression. J Neurosurg 2014;120(5):1048\u0026ndash;54. (In eng). DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3171/2014.1.Jns131410\u003c/span\u003e\u003cspan address=\"10.3171/2014.1.Jns131410\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKatusic S, Beard CM, Bergstralh E, Kurland LT. Incidence and clinical features of trigeminal neuralgia, Rochester, Minnesota, 1945\u0026ndash;1984. Ann Neurol 1990;27(1):89\u0026ndash;95. (In eng). DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/ana.410270114\u003c/span\u003e\u003cspan address=\"10.1002/ana.410270114\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eReinard K, Nerenz DR, Basheer A, et al. Racial disparities in the diagnosis and management of trigeminal neuralgia. J Neurosurg 2017;126(2):368\u0026ndash;374. (In eng). DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3171/2015.11.Jns151177\u003c/span\u003e\u003cspan address=\"10.3171/2015.11.Jns151177\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSiqueira SR, Teixeira MJ, Siqueira JT. Clinical characteristics of patients with trigeminal neuralgia referred to neurosurgery. Eur J Dent 2009;3(3):207\u0026ndash;12. (In eng).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFrazier CH. OPERATION FOR THE RADICAL CURE OF TRIGEMINAL NEURALGIA: ANALYSIS OF FIVE HUNDRED CASES. Ann Surg 1928;88(3):534\u0026ndash;47. (In eng). DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1097/00000658-192809000-00021\u003c/span\u003e\u003cspan address=\"10.1097/00000658-192809000-00021\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKERR FWL. The Etiology of Trigeminal Neuralgia. Archives of Neurology 1963;8(1):15\u0026ndash;25. DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1001/archneur.1963.00460010031003\u003c/span\u003e\u003cspan address=\"10.1001/archneur.1963.00460010031003\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eM\u0026uuml;ller HR, L\u0026eacute;vy A. Zur Pathogenese der Trigeminusneuralgie. Acta Neurochirurgica 1963;11(3):385\u0026ndash;397. DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/BF01404416\u003c/span\u003e\u003cspan address=\"10.1007/BF01404416\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBjerrum J, Thornval G. Roentgenographic findings in trigeminal neuralgia. Acta radiol 1959;51(4):289\u0026ndash;96. (In eng). DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3109/00016925909171547\u003c/span\u003e\u003cspan address=\"10.3109/00016925909171547\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSmith DG, Mumford JM. Petrous angle and trigeminal neuralgia. Pain 1980;8(3):269\u0026ndash;277. (In eng). DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/0304-3959(80)90072-x\u003c/span\u003e\u003cspan address=\"10.1016/0304-3959(80)90072-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKundak\u0026ccedil;ı YE, Doğan N, Kara İ, \u0026Ouml;zt\u0026uuml;rk M, Fazlıoğulları Z, Karabulut AK. Morphometric examination of trigeminal nerve and its adjacent structures in patients with trigeminal neuralgia: a case-control study. Turk J Med Sci 2022;52(5):1627\u0026ndash;1638. (In eng). DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.55730/1300-0144.5503\u003c/span\u003e\u003cspan address=\"10.55730/1300-0144.5503\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eParise M, Acioly MA, Ribeiro CT, Vincent M, Gasparetto EL. The role of the cerebellopontine angle cistern area and trigeminal nerve length in the pathogenesis of trigeminal neuralgia: a prospective case-control study. Acta Neurochir (Wien) 2013;155(5):863\u0026ndash;8. (In eng). DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00701-012-1573-0\u003c/span\u003e\u003cspan address=\"10.1007/s00701-012-1573-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim T, Heo J, Jang DK, et al. Machine learning for detecting moyamoya disease in plain skull radiography using a convolutional neural network. EBioMedicine 2019;40:636\u0026ndash;642. (In eng). DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ebiom.2018.12.043\u003c/span\u003e\u003cspan address=\"10.1016/j.ebiom.2018.12.043\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCruccu G, Finnerup NB, Jensen TS, et al. Trigeminal neuralgia: New classification and diagnostic grading for practice and research. Neurology 2016;87(2):220\u0026ndash;8. (In eng). DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1212/WNL.0000000000002840\u003c/span\u003e\u003cspan address=\"10.1212/WNL.0000000000002840\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStuart EA, King G, Imai K, Ho D. MatchIt: nonparametric preprocessing for parametric causal inference. Journal of statistical software 2011.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang G, Liu Z, Van Der Maaten L, Weinberger KQ. Densely connected convolutional networks. Proceedings of the IEEE conference on computer vision and pattern recognition2017:4700\u0026ndash;4708.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRussakovsky O, Deng J, Su H, et al. Imagenet large scale visual recognition challenge. International journal of computer vision 2015;115(3):211\u0026ndash;252.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRajpurkar P, Irvin J, Zhu K, et al. Chexnet: Radiologist-level pneumonia detection on chest x-rays with deep learning. arXiv preprint arXiv:171105225 2017.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePizer SM, Amburn EP, Austin JD, et al. Adaptive histogram equalization and its variations. Computer vision, graphics, and image processing 1987;39(3):355\u0026ndash;368.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSelvaraju RR, Das A, Vedantam R, Cogswell M, Parikh D, Batra D. Grad-CAM: Why did you say that? arXiv preprint arXiv:161107450 2016.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eThe International Classification of Headache Disorders, 3rd edition (beta version). Cephalalgia 2013;33(9):629\u0026ndash;808. (In eng). DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1177/0333102413485658\u003c/span\u003e\u003cspan address=\"10.1177/0333102413485658\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAziz Q, Giamberardino MA, Barke A, et al. The IASP classification of chronic pain for ICD-11: chronic secondary visceral pain. Pain 2019;160(1):69\u0026ndash;76. (In eng). DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1097/j.pain.0000000000001362\u003c/span\u003e\u003cspan address=\"10.1097/j.pain.0000000000001362\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCheraghi-Sohi S, Holland F, Reeves D, et al. The incidence of diagnostic errors in UK primary care and implications for health care, research, and medical education: a retrospective record analysis of missed diagnostic opportunities. British Journal of General Practice 2018;68:bjgp18X696857. DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3399/bjgp18X696857\u003c/span\u003e\u003cspan address=\"10.3399/bjgp18X696857\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evon Eckardstein KL, Keil M, Rohde V. Unnecessary dental procedures as a consequence of trigeminal neuralgia. Neurosurg Rev 2015;38(2):355\u0026ndash;60; discussion 360. (In eng). DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s10143-014-0591-1\u003c/span\u003e\u003cspan address=\"10.1007/s10143-014-0591-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSletteb\u0026oslash; H. Is this really trigeminal neuralgia? Diagnostic re-evaluation of patients referred for neurosurgery. Scand J Pain 2021;21(4):788\u0026ndash;793. (In eng). DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1515/sjpain-2021-0045\u003c/span\u003e\u003cspan address=\"10.1515/sjpain-2021-0045\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMcCartney S, Weltin M, Burchiel KJ. Use of an artificial neural network for diagnosis of facial pain syndromes: an update. Stereotact Funct Neurosurg 2014;92(1):44\u0026ndash;52. (In eng). DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1159/000353188\u003c/span\u003e\u003cspan address=\"10.1159/000353188\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBrinzeu A, Dumot C, Sindou M. Role of the petrous ridge and angulation of the trigeminal nerve in the pathogenesis of trigeminal neuralgia, with implications for microvascular decompression. Acta Neurochir (Wien) 2018;160(5):971\u0026ndash;976. (In eng). DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00701-018-3468-1\u003c/span\u003e\u003cspan address=\"10.1007/s00701-018-3468-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJannetta PJ. Vascular compression is the cause of trigeminal neuralgia. APS Journal 1993;2(4):217\u0026ndash;227. DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/S1058-9139(05)80246-3\u003c/span\u003e\u003cspan address=\"10.1016/S1058-9139(05)80246-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSindou M, Howeidy T, Acevedo G. Anatomical observations during microvascular decompression for idiopathic trigeminal neuralgia (with correlations between topography of pain and site of the neurovascular conflict). Prospective study in a series of 579 patients. Acta Neurochir (Wien) 2002;144(1):1\u0026ndash;12; discussion 12\u0026thinsp;\u0026ndash;\u0026thinsp;3. (In eng). DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s701-002-8269-4\u003c/span\u003e\u003cspan address=\"10.1007/s701-002-8269-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIshikawa M, Nishi S, Aoki T, et al. Operative findings in cases of trigeminal neuralgia without vascular compression: proposal of a different mechanism. J Clin Neurosci 2002;9(2):200\u0026ndash;4. (In eng). DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1054/jocn.2001.0922\u003c/span\u003e\u003cspan address=\"10.1054/jocn.2001.0922\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eParise M, Ribeiro CT, Vincent M, Gasparetto E. Small cerebellopontine angle cisterns inpatients with trigeminal neuralgia. J Neurosurg 2010;112(4):900\u0026ndash;1; author reply 901. (In eng). DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3171/2009.10.Jns091517\u003c/span\u003e\u003cspan address=\"10.3171/2009.10.Jns091517\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKuroe K, Rosas A, Molleson T. Variation in the cranial base orientation and facial skeleton in dry skulls sampled from three major populations. Eur J Orthod 2004;26(2):201\u0026ndash;7. (In eng). DOI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/ejo/26.2.201\u003c/span\u003e\u003cspan address=\"10.1093/ejo/26.2.201\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Trigeminal neuralgia, Deep learning, Convolutional neural networks, Skull X-ray","lastPublishedDoi":"10.21203/rs.3.rs-4775021/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4775021/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThis study aimed to determine whether trigeminal neuralgia can be diagnosed using convolutional neural networks (CNNs) based on plain X-ray skull images. To this end, 166 skull images of patients aged\u0026thinsp;\u0026gt;\u0026thinsp;16 years with trigeminal neuralgia diagnoses were compiled into a labeled trigeminal neuralgia dataset and 498 skull images of patients with unruptured intracranial aneurysms were compiled into a labeled control dataset. The images were partitioned into training, validation, and test datasets in a 6:2:2 ratio using random permutation. The accuracy and area under the receiver-operating characteristic (AUROC) curve were used to evaluate the classifier performance. Gradient-weighted class activation mapping was employed to identify the focal areas of attention. External validation was performed using a dataset obtained from another institution. We observed an overall accuracy of 87.2%, sensitivity and specificity of 0.72 and 0.91, respectively, and AUROC of 0.90 on the test dataset. In most cases, trigeminal neuralgia was predicted by observing the sphenoid body and clivus. The overall accuracy on the external test dataset was 71.0%, indicating the promise of deep learning-based models in distinguishing between X-ray skull images of patients with trigeminal neuralgia and control individuals. This is expected to serve as a useful screening tool after further development.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e","manuscriptTitle":"Diagnosis of trigeminal neuralgia based on plain skull radiography using convolutional neural network","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-08-28 02:51:56","doi":"10.21203/rs.3.rs-4775021/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2024-10-14T04:52:14+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-10-10T12:08:11+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-09-23T10:59:51+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"97215050090473011635082429597635335924","date":"2024-09-18T14:18:13+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"173332819531278110389745793661431873201","date":"2024-09-18T14:11:01+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-09-16T14:02:06+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-09-16T12:56:54+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2024-07-28T16:08:39+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-07-25T12:40:59+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2024-07-21T02:42:07+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"af66c133-29df-4aab-ae34-8ccf8140bb10","owner":[],"postedDate":"August 28th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":36320738,"name":"Health sciences/Anatomy"},{"id":36320739,"name":"Health sciences/Neurology"},{"id":36320740,"name":"Health sciences/Signs and symptoms"}],"tags":[],"updatedAt":"2025-06-02T16:03:13+00:00","versionOfRecord":{"articleIdentity":"rs-4775021","link":"https://doi.org/10.1038/s41598-025-03254-7","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2025-05-29 15:57:52","publishedOnDateReadable":"May 29th, 2025"},"versionCreatedAt":"2024-08-28 02:51:56","video":"","vorDoi":"10.1038/s41598-025-03254-7","vorDoiUrl":"https://doi.org/10.1038/s41598-025-03254-7","workflowStages":[]},"version":"v1","identity":"rs-4775021","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4775021","identity":"rs-4775021","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}