Establishment and validation of a model for predicting postoperative recurrence of sinonasal tumors based on multimodality MRI parameters

Introduction

Sinonasal tumors are a rare disease with a global annual incidence rate of less than 1 per 100,000. They account for less than 1% of all malignant tumors and less than 3% of head and neck cancers. These tumors most frequently occur in individuals aged 50–70 years, with the incidence rate in men being approximately twice that in women (2:1) (1-3). Its histological characteristics and clinical manifestations are diverse and should be regarded as an independent entity rather than a subtype of head and neck cancer (4). The 5-year overall survival rates for early (T1–2) and advanced patients are 60% and 20% (5), respectively. Most new cases are locally advanced or metastatic cancers that are difficult to receive radical treatment. Surgery is the main treatment (6), but the high recurrence rate after surgery leads to many difficulties in treatment (7). To help doctors predict better postoperative recurrence of sinonasal tumors, a model is necessary.

Previousresearch mainly explored the predictive value of clinical and laboratory indicators for postoperative recurrence (8). This approach has the advantages of being easily accessible, cost-effective, and having a wide existing research foundation (9). However, it is limited by insufficient accuracy and specificity, significant individual variability, and difficulty in reflecting dynamic changes (10). In contrast, imaging technology can provide more comprehensive information on tissue and anatomical structures, aiding in personalized treatment and early detection of potential recurrence (11). Therefore, combining traditional clinical indicators with image analysis is expected to enhance the accuracy and reliability of postoperative recurrence predictions.

Magnetic resonance imaging (MRI) is a medical imaging technique that uses strong magnetic fields and radiofrequency pulses to obtain detailed images of the internal structures of the body, which can provide high-resolution images to help doctors diagnose and evaluate various medical issues, such as brain disorders, spinal cord injuries, joint injuries, and cancer (12). Unlike X-rays and computed tomography (CT) scans, MRI does not use ionizing radiation and is generally considered a safe imaging method (13). By analyzing the response of hydrogen atoms in the body, MRI can generate detailed images of different tissues, making it particularly suitable for observing soft tissues (13).

Ramkumar et al. (14) applied machine learning to distinguish inverted papilloma from sinus squamous cell carcinoma using MRI images. In addition, Gu et al. (15) applied machine learning to distinguish inverted papilloma (IP) from malignant sinus tumors via MRI images. In summary, functional MRI parameters and radiomic analysis have shown promising prognostic value in predicting treatment response and recurrence in the more common head and neck cancers. However, their application specifically to sinonasal tumors remains underexplored.

This study aims to develop a predictive model for postoperative recurrence of sinonasal tumors utilizing multimodal MRI parameters. This innovative approach transcends the limitations of prior research by integrating image-based features into the prediction framework, thereby addressing critical research gaps and offering fresh perspectives and directions for the field. It also enriches the theoretical underpinnings of the subject. Multimodal MRI parameters provide a multi-dimensional view of tumor characteristics, thereby enhancing prediction accuracy and furnishing a reliable foundation for clinical decision-making. Consequently, clinicians can tailor individualized treatment plans, optimize the allocation of medical resources, and prioritize high-risk patients. Moreover, accurate predictions facilitate early intervention for potential relapses, thereby improving treatment efficacy, prolonging survival, and enhancing long-term prognosis and quality of life for patients. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1065/rc).

Methods

Research subjects

A total of 180 patients with sinonasal tumors admitted to Wudang Hospital Affiliated to Guizhou Medical University from January 2021 to December 2023 were retrospectively selected and randomized into the training group (n=135) and the test group (n=45) in a 3:1 ratio. Inclusion criteria: (I) all patients met the surgical indications for sinonasal tumors (16); (II) the patient is ≤75 years old; (III) the patient had no previous history of radiotherapy or chemotherapy. Exclusion criteria: (I) patients with other malignant tumors; (II) presence of mental disorder; (III) the patient has severe brain trauma; (IV) patients with contrast agent allergies; (V) patients with MRI contraindications.

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of Wudang Hospital Affiliated to Guizhou Medical University (Approval No. GMUWD2021011061003). Informed consent was waived in this retrospective study.

Grouping

Patients in the modeling group were divided into two groups according to whether they recurred after operation for 6 months: postoperative recurrence group (n=50) and non-postoperative recurrence group (n=85). Judgment criteria (17): (I) observation of general symptoms: recurrent sinus cancer is often manifested as nasal swelling and nasal congestion, which are similar to the symptoms at the initial onset. If the tumor has a high vascularity, nosebleeds may occur. Visual acuity and blurred vision are common at recurrence, which may be related to tumor invasion of the associated nerves. In the middle and advanced stages, patients may also experience local numbness and pain as a manifestation of tumor spread to surrounding tissues. (II) Fiberoptic nasopharyngoscopy: perform fiberoptic nasopharyngoscopy to observe whether there are signs of tumor recurrence in the nasopharynx.

Detection methods

During MRI examinations using the 3.0T MAGNETOM Trio Tim system (Siemens, Erlangen, Germany) and a dedicated head and neck surface coil, all patients were first instructed to remove any metallic objects and underwent thorough screening for contraindications (e.g., incompatible implants, severe renal impairment) and allergy history (particularly to gadolinium-based contrast agents). Patients were then positioned supine and comfortably immobilized to minimize motion artifacts throughout the examination. The imaging protocol included high-resolution three-plane (axial, coronal, sagittal) T2-weighted fast spin-echo (SE) localizer images to accurately plan subsequent sequences. Routine anatomical sequences covered the entire sinonasal tract and skull base, specifically including coronal and axial T2-weighted (T2W) imaging using fat-suppressed turbo spin-echo (TSE) sequences with typical parameters of repetition time (TR)/echo time (TE) =3,000–4,000/85–100 ms, slice thickness =3 mm, interslice gap =0.5 mm, field of view (FOV) =200×200 mm², matrix size =320×320; axial T1-weighted (T1W) imaging was performed pre- and post-contrast using SE or TSE sequences with parameters of TR/TE =500–600/8–10 ms, slice thickness =3 mm, interslice gap =0.5 mm, FOV =200×200 mm², matrix size =256×256. Dynamic contrast-enhanced (DCE) MRI was acquired using a T1-weighted 3D spoiled gradient echo sequence with high temporal resolution, key parameters being TR/TE =5.1/0.9 ms, flip angle =15°, FOV =240×240 mm², matrix size =192×192, slice thickness =3.0 mm, temporal resolution =9–10 seconds. A T1 mapping was performed prior to contrast injection, acquiring two sets of images at flip angles of 2° and 15° to calculate baseline T1. After acquiring the first five dynamic phases, gadobutrol (Gadovist, Bayer Pharma, Berlin, Germany; 0.2 mL/kg) was injected intravenously at a flow rate of 3 mL/s using a power injector, followed by a 20 mL saline flush. A total of 38 dynamic phases were acquired continuously, capturing contrast uptake and washout kinetics over approximately 6 minutes. DCE-MRI data were transferred to a dedicated post-processing workstation for quantitative analysis using syngo.via. The processing involved motion correction by automatic or manual registration to correct patient motion during dynamic acquisition; arterial input function (AIF) selection, where the radiologist manually placed a small region of interest (ROI) in a major feeding artery (such as the external carotid artery or its branch) on parametric maps to derive the AIF; ROI analysis involved careful delineation of three separate ROIs on the axial slice of the solid, intensely enhancing portion of the tumor on the parametric map, avoiding areas of necrosis, cysts, hemorrhage, and large vessels; model fitting applied the extended Tofts pharmacokinetic model at each voxel to generate quantitative parametric maps. Finally, quantitative analysis of the acquired DCE-MRI images was performed, calculating parameters such as the rate constant (Kep), volume transfer constant (Ktrans), extravascular extracellular space volume fraction (Ve), and plasma volume fraction (Vp).

All quantitative DCE-MRI parameter maps (Ktrans, Kep, Ve, Vp) were generated using the Philips IntelliSpace Portal with an extended Tofts model. Two radiologists with several years of experience in head and neck imaging manually performed the ROI segmentation on axial contrast-enhanced T1-weighted images, blinded to clinical outcomes and recurrence status. ROIs were carefully delineated along the inner margin of the enhancing tumor border, avoiding areas of necrosis, cysts, significant hemorrhage, large feeding vessels, and prominent artifacts to maintain measurement accuracy. For each patient, three non-overlapping ROIs were placed on three consecutive slices displaying the largest cross-sectional area of the tumor, with each ROI size ranging from 15–25 mm². The final values for each parameter were calculated as the mean of these three ROIs to enhance reliability and account for intratumoral heterogeneity. For smaller tumors, where placing three ROIs was not feasible, a single ROI covering the entire solid tumor area on the largest slice was used, aiming to maximize sampling of tumor heterogeneity while minimizing partial volume effects.

Assessment of signal unevenness

Inhomogeneous signal refers to the uneven distribution of internal signal intensity within tumor tissue across different MRI sequences. In other words, within the same tumor area, there are notable variations in signal intensity, with some regions showing high signals and others showing low signals, or the signal changes are inconsistent across different sequences.

The evaluation of ‘signal unevenness’ was performed qualitatively on T2-weighted and contrast-enhanced T1-weighted images. It was defined as the presence of a non-uniform pattern of signal intensity within the tumor mass, characterized by a mix of hyperintense and hypointense areas relative to the overall tumor signal. Specifically, a tumor was classified as having an ‘uneven signal’ if it exhibited any of the following: (I) intratumoral cysts or necrosis (well-defined areas of fluid-like T2 hyperintensity with no post-contrast enhancement); (II) intratumoral hemorrhage (variable signal on T1/T2 but often T1 hyperintense); or (III) marked textural heterogeneity (a conspicuous, patchy distribution of signal intensities without the well-defined characteristics of cysts or hemorrhage). Tumors with a homogeneous signal or only minimal, faint heterogeneity were classified as having an ‘even signal’. All assessments were made by two radiologists in consensus.

Data collection

General data, including age, gender, and body mass index (BMI), were collected using the electronic medical record system and inquiry methods. MRI-related data were obtained through MRI examinations.

Statistical analysis

The collected experimental data were analyzed using SPSS 27.0. Non-normally distributed measurement data are presented as median and interquartile range (IQR), and inter-group comparisons were performed using the rank-sum test. Normally distributed measurement data are expressed as mean ± standard deviation, and inter-group comparisons were conducted using the independent-samples t-test. For multiple group comparisons, the F-test was used. Enumeration data are presented as the number of cases or rates, and comparisons were made using the χ² test or Fisher’s exact test, as appropriate. Univariate and multivariate logistic regression analyses were employed to identify the influencing factors of postoperative recurrence in sinonasal tumors and to establish a prediction model. The receiver operating characteristic (ROC) curve was used to evaluate the predictive performance of the model, with statistical significance set at P<0.05.

Results

Baseline data

The training group consisted of 69 males and 66 females, with an average age of 54.48±6.48 years and a mean BMI of 22.71±2.15 kg/m². The pathological types included 74 cases of squamous cell carcinoma, 36 cases received postoperative chemotherapy, 20 cases were T1–2, and 115 cases were T3–4. The test group included 25 males and 20 females, with a mean age of 54.73±6.35 years and a mean BMI of 22.97±2.22 kg/m². The pathological types in the test group included 23 cases of squamous cell carcinoma, 11 cases received postoperative chemotherapy, 10 cases were T1–2, and 35 cases were T3–4. The recurrence rates for the training and test groups were 11.11% and 13.33%, respectively. There was no statistically significant difference between the training and test groups (P>0.05), as shown in Table 1.

Factors influencing single factor analysis

Comparison between the two groups in the training group exhibited no significant differences in terms of age, gender, BMI, T1 hypersignal, T2 hyposignal, tumor size, margin, mucoid change, enhancement mode, enhancement depth, operation time, pathological type, postoperative chemotherapy, and stage (P>0.05). There were statistical significances in terms of uneven signal, Vp, Ve, septum, and cutting margin (P<0.05), as shown in Table 2.

Binary logistic regression analysis influencing factors

Using signal unevenness, Vp, Ve, separation, and resection margin as independent variables, and after assigning values, analysis was performed with whether recurrence after surgery was the dependent variable (recurrence =1, non-recurrence =0). The results of binary Logistics regression analysis showed that signal unevenness [Exp(B) =7.691, 95% confidence interval (CI): 1.153–2.288], Vp [Exp(B) =3.236, 95% CI: 1.609–6.511], Ve [Exp(B) =0.579, 95% CI: 0.437–0.766], and resection margin [Exp(B) =8.102, 95% CI: 1.260–5.107],were independent predictors of postoperative recurrence (P<0.05), see Table 3.

Construction of the predictive models

According to the results of Logistic regression analysis, signal unevenness, Vp, Ve, cutting margin were analyzed (named X1, X2, X3, and X4 respectively) were included in the constructed prediction model, and the joint detection factor model expression was Logit(P) =−4.71.2 + (2.040X1) + (1.174X2) + (−0.547X3) + (2.092X4), the slopes of the calibration curves in the training under construction and test groups are straight lines approximating 1, indicating that the risks predicted by the model are in good agreement with the actual risks (Figure 1A,1B).

Figure 1 Calibration curves of the prediction model for postoperative recurrence of sinonasal tumors. (A) Calibration curve in the training set. (B) Calibration curve in the validation set. The diagonal dashed line represents perfect prediction, and the solid line represents the model's performance. Closer alignment of the solid line to the diagonal indicates better calibration. MRI, magnetic resonance imaging.

ROC analysis: application value of prediction model in predicting postoperative recurrence of nasal and sinus tumors

ROC analysis results showed that the area under the model’s predictive curve for postoperative recurrence in patients with nasal and sinus tumors in the training set was 0.955 (95% CI: 0.9135–0.9637), and the standard error was 0.024, the best cutoff value is 0.69, the sensitivity at this time is 95.49%, and the specificity is 73.48%, as shown in Figure 2; the area under the curve of the model for postoperative recurrence of patients with nasal and sinus tumors in the validation set is 0.831 (95% CI: 0.7357–0.9248), and the standard error is 0.048, the best cutoff value is 0.29, at this time the sensitivity is 67.48%, and the specificity is 61.37%, as shown in Figure 2.

Figure 2 ROC curves of the prediction model for postoperative recurrence in the training set (A) and the validation set (B). AUC, area under the curve; ROC, receiver operating characteristic.

Clinical benefit analysis of the predictive models

In Figure 3, the horizontal line along the x-axis represents a scenario where no patients experience recurrence or receive intervention, resulting in a net benefit rate of 0. The gray diagonal line represents a scenario where all patients experience recurrence and receive intervention. The farther the red curve is from these two lines, the higher the net benefit. It is evident from the figure that when the predicted high-risk threshold is between 0.08 and 0.98, the model demonstrates high clinical application value.

Figure 3 Decision curve analysis (DCA) of the prediction model. The red curve represents the predictive model. The gray diagonal line indicates the “all patients receive intervention” scenario, and the horizontal black line indicates the “no patients receive intervention” scenario. The model demonstrates a higher net benefit across a wide range of threshold probabilities, indicating favorable clinical applicability.

Discussion

The prediction of postoperative recurrence rate for sinonasal tumors, especially sinonasal squamous cell carcinoma (SNSCC), a head and neck malignancy, has always been the focus of clinical attention (18). With the continuous advancements in medical imaging technology, multimodality MRI is playing an increasingly important role in tumor diagnosis, treatment, and recurrence prediction. This is because it can provide comprehensive information on various aspects such as tumor structure, function, and metabolism (19). This study successfully established and validated a model for predicting postoperative recurrence of sinus tumors based on multimodal MRI parameters. This model provides clinicians with a more accurate tool for assessing recurrence risk and guides the formulation of individualized treatment plans.

Multiple factors associated with postoperative recurrence of sinonasal tumors are identified by univariate and multivariate logistic regression analyses. Among them, signal unevenness, margin, separation, Vp, and Ve are independent influencing factors for postoperative recurrence of patients with nasal and sinus tumors. The findings of these parameters not only reveal the close relationship between tumor recurrence and MRI features, but also provide important predictive variables for subsequent modeling. Based on the above influencing factors, a prediction model for postoperative recurrence of sinonasal tumors was established. The prediction model established in this study has high clinical application value. Through ROC analysis, the results showed that the area under the curve of the model for predicting postoperative recurrence in patients with sinonasal tumors was 0.955 in the training set, and 0.831 in the validation set. The model shows good prediction performance in both the training set and the validation set, and the AUC is high, demonstrating that the model has high sensitivity and specificity. In addition, the slope of the calibration curve is close to 1, further validating the consistency between model-predicted risk and actual risk. First, it can provide clinicians with a more accurate recurrence risk assessment tool to help them better understand the patient’s condition and prognosis, thus formulating a more individualized treatment regimen (20,21). Secondly, the model can support clinicians in surgical decision-making, including the determination of whether extended resection or adjuvant therapy is warranted. Furthermore, this model contributes to enhanced treatment satisfaction and quality of life among patients. By enabling early prediction of recurrence risk, it facilitates timely preventive interventions or earlier initiation of treatment, thereby mitigating the physical and psychological burden associated with disease recurrence (22). As a non-invasive and radiation-free medical imaging technology, MRI plays an important role in tumor diagnosis, treatment, and recurrence prediction (23). In this study, multiple MRI parameters were selected as predictive variables and a prediction model based on these parameters was successfully established. These parameters are derived not only from established literature and clinical expertise but also incorporate comprehensive data on tumor morphology, functional characteristics, and metabolic activity. By integrating these multidimensional features, the model enables more precise tumor characterization and, consequently, improves the accuracy of recurrence risk prediction.

Although this study successfully established a model for predicting postoperative recurrence of sinonasal tumors based on multimodality MRI parameters, there are still focal regions that can be optimized and improved. First, the sample size can be further increased to improve the stability and generalization ability of the model (24). Secondly, more MRI parameters or other imaging features can be introduced to enrich the predictive variables of the model and improve the prediction accuracy. In addition, other clinical information, or data such as molecular markers can be combined to build a more comprehensive prediction model, which may facilitate the development of a more comprehensive predictive framework (25). With the continuous progress of medical imaging technology and data analysis methods, tumor recurrence prediction models based on multimodal MRI parameters will be more widely used in the future (26,27). First, the model can be applied to predict various types of tumor recurrence and provide clinicians with a more accurate assessment tool (28). Secondly, the model can also be combined with other medical information systems to integrate and share patient information and improve the efficiency and quality of medical services (29). Last but not least, regarding modeling techniques, we employed binary logistic regression for its interpretability, simplicity, and suitability for this preliminary investigation.

On the other hand, this choice may not capture more complex, non-linear relationships between the predictors and outcome. Future studies could leverage advanced machine learning algorithms [e.g., random forest, support vector machines, or eXtreme Gradient Boosting (XGBoost)] to explore these potential complex interactions. These methods might enhance predictive performance and provide deeper insights into the underlying data structure. The integration of such techniques with our established MRI parameters represents a promising avenue for building more powerful and accurate predictive models.

An important consideration for the future clinical translation of our model is the harmonization of MRI acquisition protocols across different institutions and scanner platforms. Our current model was developed and validated on a single scanner using a specific protocol, which limits its generalizability. Variations in field strength, sequence parameters, and reconstruction algorithms can significantly affect quantitative MRI parameters like Vp and Ve, leading to potential performance degradation in external settings. To address this, future multi-center validation studies must implement protocol standardization strategies. This could involve the use of standardized imaging phantoms for cross-calibration, adherence to consensus guidelines [e.g., those from the Quantitative Imaging Biomarkers Alliance (QIBA)], and the application of advanced statistical harmonization techniques (such as ComBat) to minimize site-specific biases before model deployment. This step is crucial for ensuring that our predictive model yields consistent and reliable results across diverse clinical environments.

Our study incorporated the qualitative assessment of ‘signal unevenness’, which is inherently semi-subjective. Although this feature was a significant predictor in our model and was defined using standardized criteria, the lack of a formal interobserver agreement analysis remains a limitation. This introduces potential variability, which may affect the generalizability and reproducibility of this specific parameter. To overcome the limitations of subjective visual assessment, our future work will prioritize the extraction of high-dimensional, quantitative radiomic features from preoperative MRI. These features can objectively capture tumor heterogeneity, texture, and intensity patterns with superior reproducibility. Integrating such automated, data-driven biomarkers with clinical and DCE-MRI parameters holds the greatest promise for building robust and generalizable predictive models.

Furthermore, we observed a significant decline in the model’s predictive performance when transitioning from the training set to the validation set. This discrepancy serves as a clear indication of potential overfitting, a prevalent issue in radiomic and predictive modeling studies, particularly when dealing with limited sample sizes. While the model exhibited robust discriminatory power within the training cohort, its performance was likely inflated by learning dataset-specific noise and patterns, thereby compromising its generalizability to the independent validation set. This overfitting is likely due to the relatively small size of our single-center cohort and the model’s complexity relative to the number of outcome events. As a result, the current model, in its present state, is not suitable for clinical application. Addressing how to effectively adapt this model for clinical diagnosis will be a central focus of our subsequent research endeavors.

A key limitation of the present study is the reliance on manual segmentation for parameter extraction, which may introduce observer variability. To address this, we plan to integrate semi-automated and AI-assisted contouring tools in subsequent work. The adoption of these advanced methodologies, which represent the current best practice in radiomics, will minimize manual intervention, maximize reproducibility, and further strengthen the reliability of the imaging biomarkers we have identified. This evolution towards fully automated pipelines is essential for developing robust, generalizable models ready for future clinical translation.

Furthermore, our model employs a binary endpoint (recurrence within 6 months), which does not incorporate time-to-event data. While this provides a clear initial framework, it fails to capture the temporal dimension of recurrence, such as distinguishing between early and late recurrences, which is critical for planning personalized follow-up schedules and understanding the full spectrum of disease aggressiveness. In future research, we will further improve this shortcoming.

Conclusions

In conclusion, we have developed and externally validated an MRI-based model that effectively stratifies the risk of recurrence following resection of sinonasal tumors. This model equips clinicians with a precise tool for assessing relapse risk, thereby facilitating the formulation of personalized treatment regimens and enhancing patients’ treatment efficacy and quality of life. Moving forward, we are committed to further refining and optimizing this model, as well as exploring its broader application potential.

Acknowledgments

The authors express their appreciation to staff in their hospital, for their technical assistance.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1065/rc

Data Sharing Statement: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1065/dss

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1065/prf

Funding: None.

Conflicts of Interest: Both authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1065/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of Wudang Hospital Affiliated to Guizhou Medical University (Approval No. GMUWD2021011061003). Informed consent was waived in this retrospective study.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Abdelmeguid AS, Bell D, Hanna EY. Sinonasal Undifferentiated Carcinoma. Curr Oncol Rep 2019;21:26. [Crossref] [PubMed]
2. Bracigliano A, Tatangelo F, Perri F, et al. Malignant Sinonasal Tumors: Update on Histological and Clinical Management. Curr Oncol 2021;28:2420-38. [Crossref] [PubMed]
3. Thariat J, Moya Plana A, Verillaud B, et al. Diagnosis, prognosis and treatment of sinonasal carcinomas (excluding melanomas, sarcomas and lymphomas). Bull Cancer 2020;107:601-11. [Crossref] [PubMed]
4. Abad J, Llop E, Arias-Loste MT, et al. Endoscopic Sleeve Gastroplasty Plus Lifestyle Intervention in Patients With Metabolic Dysfunction-associated Steatohepatitis: A Multicenter, Sham-controlled, Randomized Trial. Clin Gastroenterol Hepatol 2025;23:1556-1566.e3. [Crossref] [PubMed]
5. Birkenbeuel JL, Goshtasbi K, Adappa ND, et al. Recurrence rates of de-novo versus inverted papilloma-transformed sinonasal squamous cell carcinoma: a meta-analysis. Rhinology 2022;60:402-10. [Crossref] [PubMed]
6. Lombardo N, Della Corte M, Pelaia C, et al. Primary Mucosal Melanoma Presenting with a Unilateral Nasal Obstruction of the Left Inferior Turbinate. Medicina (Kaunas) 2021;57:359. [Crossref] [PubMed]
7. Tsuda T, Hosokawa K, Fujii S, et al. A Case of Sinonasal Inverted Papilloma Suspected as Postoperative Recurrence of Eosinophilic Chronic Rhinosinusitis. Cureus 2024;16:e69971. [Crossref] [PubMed]
8. Miao S, Cheng Y, Li Y, et al. Prediction of recurrence-free survival and risk factors of sinonasal inverted papilloma after surgery by machine learning models. Eur J Med Res 2024;29:528. [Crossref] [PubMed]
9. Kate RJ, Pearce N, Mazumdar D, et al. A continual prediction model for inpatient acute kidney injury. Comput Biol Med 2020;116:103580. [Crossref] [PubMed]
10. Vernooij LM, van Klei WA, Moons KG, et al. The comparative and added prognostic value of biomarkers to the Revised Cardiac Risk Index for preoperative prediction of major adverse cardiac events and all-cause mortality in patients who undergo noncardiac surgery. Cochrane Database Syst Rev 2021;12:CD013139. [Crossref] [PubMed]
11. Dobai A, Dembrovszky F, Vízkelety T, et al. MRI compatibility of orthodontic brackets and wires: systematic review article. BMC Oral Health 2022;22:298. [Crossref] [PubMed]
12. Wekking D, Porcu M, De Silva P, et al. Breast MRI: Clinical Indications, Recommendations, and Future Applications in Breast Cancer Diagnosis. Curr Oncol Rep 2023;25:257-67. [Crossref] [PubMed]
13. Li S, Dai Y, Chen J, et al. MRI-based habitat imaging in cancer treatment: current technology, applications, and challenges. Cancer Imaging 2024;24:107. [Crossref] [PubMed]
14. Ramkumar S, Ranjbar S, Ning S, et al. MRI-Based Texture Analysis to Differentiate Sinonasal Squamous Cell Carcinoma from Inverted Papilloma. AJNR Am J Neuroradiol 2017;38:1019-25. [Crossref] [PubMed]
15. Gu J, Yu Q, Li Q, et al. MRI radiomics-based machine learning model integrated with clinic-radiological features for preoperative differentiation of sinonasal inverted papilloma and malignant sinonasal tumors. Front Oncol 2022;12:1003639. [Crossref] [PubMed]
16. Thiel HJ, Rettinger G. Current status of detection and treatment of malignant nasal and paranasal sinus tumors. 1. Pathology, diagnosis and staging of nasal and paranasal sinus tumors. Hno 1986;34:91-5.
17. Jain R, Sankar R, Singh AB, et al. Endoscopic Management of Malignant Tumors of Paranasal Sinus and Nasal Cavity: An Institutional Experience. Indian J Otolaryngol Head Neck Surg 2022;74:564-74. [Crossref] [PubMed]
18. Kumari S, Pandey S, Verma M, et al. Clinicopathological Challenges in Tumors of the Nasal Cavity and Paranasal Sinuses: Our Experience. Cureus 2022;14:e29128. [Crossref] [PubMed]
19. Laskar SG, Pai P, Sinha S, et al. Intensity-modulated radiation therapy for nasal cavity and paranasal sinus tumors: Experience from a single institute. Head Neck 2021;43:2045-57. [Crossref] [PubMed]
20. Li G, Qiu B, Huang YX, et al. Cost-effectiveness analysis of proton beam therapy for treatment decision making in paranasal sinus and nasal cavity cancers in China. BMC Cancer 2020;20:599. [Crossref] [PubMed]
21. Li J, Li B, Xu J, et al. A retrospective review of non-intestinal-type adenocarcinoma of nasal cavity and paranasal sinus. Oncol Lett 2023;25:132. [Crossref] [PubMed]
22. Elias S, Benevides ML, Martins ALP, et al. Factors associated with post-stroke depression in the acute phase of ischemic stroke: A cross-sectional study. Clin Neurol Neurosurg 2022;223:107505. [Crossref] [PubMed]
23. Lian B, Yang Y, Zheng B, et al. Efficacy and Safety of Postoperative Adjuvant Radiation Therapy in Resected Nasal Cavity and Paranasal Sinus Mucosal Melanoma: A Combined Analysis. Int J Radiat Oncol Biol Phys 2024;120:528-36. [Crossref] [PubMed]
24. Nagano H, Matsumoto H, Miyamoto Y, et al. Adult T-cell Leukemia/Lymphoma (ATL) in the Nasal and Paranasal Cavity: Four Cases Report. Indian J Otolaryngol Head Neck Surg 2024;76:1264-71. [Crossref] [PubMed]
25. Wang Z, Qu Y, Wang K, et al. The value of preoperative radiotherapy in the treatment of locally advanced nasal cavity and paranasal sinus squamous cell carcinoma: A single institutional experience. Oral Oncol 2020;101:104512. [Crossref] [PubMed]
26. Owin N, Elsayad K, Rolf D, et al. Radiotherapy as Part of Treatment Strategies in Nasal Cavity and Paranasal Sinus Malignancies. Anticancer Res 2021;41:1587-92. [Crossref] [PubMed]
27. Saito T, Nakayama M, Ohnishi K, et al. Proton beam therapy in multimodal treatment for locally advanced squamous cell carcinoma of the nasal cavity and paranasal sinus. Radiat Oncol 2023;18:106. [Crossref] [PubMed]
28. Sharma RK, Irace AL, Schlosser RJ, et al. Conditional and Overall Disease-Specific Survival in Patients With Paranasal Sinus and Nasal Cavity Cancer: Improved Outcomes in the Endoscopic Era. Am J Rhinol Allergy 2022;36:57-64. [Crossref] [PubMed]
29. Thompson LDR, Bishop JA. Update from the 5th Edition of the World Health Organization Classification of Head and Neck Tumors: Nasal Cavity, Paranasal Sinuses and Skull Base. Head Neck Pathol 2022;16:1-18.