CT Radiomics-Clinical Model for Noninvasive Prediction of Tumor Mutation Burden and Survival in Locally Advanced Head and Neck Squamous Cell Carcinoma

CT Radiomics-Clinical Model for Noninvasive Prediction of Tumor Mutation Burden and Survival in Locally Advanced Head and Neck Squamous Cell Carcinoma | 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 Research Article CT Radiomics-Clinical Model for Noninvasive Prediction of Tumor Mutation Burden and Survival in Locally Advanced Head and Neck Squamous Cell Carcinoma Lu Li, Youjing Qiu, Yuan Qin, Jirui Xie, Jianlan Ren, Mei Feng, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6921523/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Purpose Tumor mutation burden (TMB) has emerged as a promising biomarker for predicting immunotherapy response. This study aimed to evaluate the potential of a CT-based radiomics model to predict TMB status and overall survival (OS) in patients with head and neck squamous cell carcinoma (HNSCC). Methods Somatic mutation and transcriptome data of 506 HNSCC cases were retrieved from The Cancer Genome Atlas (TCGA) to calculate TMB and identify differentially expressed genes (DEGs). Functional enrichment analysis was conducted using Gene Ontology (GO) and KEGG databases. Kaplan-Meier and Cox regression analysis was used to assess the prognostic value of TMB. A cohort of 159 patients with pre-treatment contrast-enhanced CT scan from The Cancer Imaging Archive (TCIA) was used to extract radiomics features. The dataset was split into training (n = 112) and validation (n = 47) datasets. Feature selection was performed using univariate Cox regression and LASSO, followed by multivariate Cox analysis. Predictive models (radiomics, clinical, and combined) were evaluated using receiver operating characteristic (ROC) curves and area under the curve (AUC) values. Results High TMB was associated with poor OS. Functional enrichment revealed that DEGs were enriched in immune processes and cell signaling. Notably, 16 features were selected for the TMB prediction model. The combined model achieved AUC values of 0.902 (training) and 0.669 (validation) for TMB prediction. For 1-, 3-, and 5-year OS prediction, the combined model achieved AUC values of 0.665–0.784 in the training cohort and 0.680–0.772 in the validation cohort. Conclusion The CT-derived radiomics model demonstrated potential for noninvasive prediction of TMB and survival in HNSCC, appearing advantageous for immunotherapy decision-making. enhanced-CT radiomics tumor mutational burden head and neck squamous cell carcinoma Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Introduction Head and neck squamous cell carcinoma (HNSCC) is a major cause of cancer-related mortality, with locally advanced (LA) stages presenting significant treatment challenges and poor prognosis. Despite the use of chemoradiotherapy as the standard treatment for LA-HNSCC, outcomes remain suboptimal, with a high rate of recurrence and metastasis. Recent studies, such as the JAVELIN Head and Neck 100 trial and the KEYNOTE-412 study, explored the efficacy of combining PD-L1 or PD-1 inhibitors with chemoradiotherapy but failed to show significant improvements in overall survival (OS). The KEYNOTE-412 trial indicated a trend toward improved progression-free survival (PFS) for patients with a high combined positive score (CPS ≧ 20), yet OS benefits remained limited. These findings highlight the need for more effective strategies to predict treatment response and improve survival outcomes for LA-HNSCC patients. Numerous studies have explored biomarkers predictive of response to immune checkpoint inhibitors (ICIs), including PD-L1 expression, Combined Positive Score (CPS), and tumor mutational burden (TMB) [ 7 ]. Among these, a pan-cancer analysis identified TMB—particularly clonal TMB—as a robust predictor of response to ICI therapy [ 8 ]. Whole-exome sequencing (WES) remains the gold standard for TMB measurement; however, its widespread clinical implementation is hindered by high costs, extended turnaround times, and limited accessibility. Radiomics, a high-throughput image analysis technique, enables the extraction of a large number of quantitative features from standard medical images, capturing intratumoral heterogeneity that is invisible to the naked eye. By converting biomedical images into mineable data, radiomics has shown promise in preoperative evaluation, tumor classification, prognosis assessment, and treatment response prediction [ 9 , 10 ]. Compared to WES, radiomics offers a noninvasive, cost-effective, and widely accessible alternative for evaluating tumor biology in clinical practice. Radiogenomics, the integration of radiomics and genomic data, provides a framework to infer genomic and molecular characteristics—such as gene expression profiles, methylation patterns, and TMB—from imaging features. This approach enhances clinical decision-making by linking radiological phenotypes with underlying tumor biology [ 11 – 13 ]. Importantly, radiomics-based TMB prediction may offer a clinically feasible and scalable method for immunotherapy stratification in settings where genomic testing is impractical. The present study aimed to develop a noninvasive radiomics approach to predict TMB status and survival outcomes in patients with LA-HNSCC. Genomic data were retrieved from The Cancer Genome Atlas (TCGA) to calculate TMB and analyze its associations with clinical characteristics, immune cell infiltration, and related gene expression. Radiomic features were extracted from computed tomography (CT) images of 159 patients obtained from The Cancer Imaging Archive (TCIA). Using these features, we constructed predictive models for TMB status and evaluated their association with clinical parameters and OS. This research demonstrates how radiomics can complement genomic analysis in precision oncology by providing an accessible imaging-based alternative for tumor profiling. Methods Patient cohort and data collection The study cohort consisted of locally advanced (LA) HNSCC patients. Transcriptome sequencing data for a total of 506 HNSCC cases, including LA-HNSCC patients, along with clinical and follow-up data, were downloaded from The Cancer Genome Atlas (TCGA) database ( https://portal.gdc.cancer.gov/ ). Samples missing complete clinical information, those with survival times less than 30 days, or those that were not solid tumors or lacked sequencing data were excluded from the study. For tumor staging, the patients were classified based on the AJCC staging system, focusing specifically on those categorized as locally advanced (LA-HNSCC). The TCGA whole-exome somatic mutation data were obtained in Mutation Annotation Format (MAF) and processed using the VarScan software to calculate tumor mutational burden (TMB). Gene expression profiles in fragments per kilobase of transcript per million mapped reads (FPKM) format were converted to transcripts per million (TPM) format for subsequent analyses. Clinical data included age, sex, race, smoking history, drinking history, primary tumor site, human papillomavirus (HPV) infection status, histological differentiation, American Joint Committee on Cancer (AJCC) stage, survival time, and survival status. HPV infection positivity was defined as p16 protein positivity by immunohistochemistry or HPV positivity by fluorescence in situ hybridization. Gene set variation analysis Differentially expressed genes (DEGs) were identified using the R package limma , with thresholds set at |log2 fold change| >1.0 and a false discovery rate (FDR) < 0.05. A heatmap was generated using the pheatmap package to visualize gene expression differences. Functional enrichment analyses, including Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses, were conducted using the clusterProfiler package, with significance set at p- and q-values < 0.05. Calculation of TMB value TMB was defined as the total number of somatic coding mutations, including base substitutions, insertions, and deletions, detected per megabase of tumor DNA. Based on a previous study [ 15 ], the TMB value for each sample was calculated as the total number of mutations divided by the length of the human exonic region (38 Mb). Synonymous mutations were excluded from the TMB calculation. Mutation data for HNSCC samples, downloaded from TCGA, were analyzed using the R package maftools . The optimal TMB cut-off value was determined based on the area under the receiver operating characteristic (ROC) curve (AUC) for survival prediction using X-tile software. Patients were stratified into low- and high-TMB groups according to the median TMB value (4.2 mutations/Mb). TMB values were then merged with corresponding survival information by matching patient ID numbers. Kaplan–Meier survival analysis was performed using R packages to evaluate the association between TMB and prognosis in HNSCC. Additionally, the relationship between TMB values and clinical characteristics was assessed. Image acquisition and segmentation A total of 159 LA-HNSCC cases with enhanced CT images in the arterial phase were obtained from The Cancer Imaging Archive (TCIA) database ( https://www.cancerimagingarchive.net/ ), specifically from the Head and Neck Squamous Cell Carcinoma (HNSCC) Imaging Dataset. Only arterial phase CT images were included in the analysis to ensure consistency in the vascular contrast enhancement, which is typically most relevant for assessing tumor characteristics in HNSCC. The decision to use only the arterial phase images was based on prior studies indicating that this phase provides the most accurate representation of tumor vascularization and tumor-tissue differentiation, which are crucial for radiomics feature extraction and TMB prediction. Other CT phases, such as venous or delayed phases, were not considered due to their lesser ability to delineate tumor boundaries and vascular features in the context of HNSCC. The images were processed and analyzed to create a radiomics model. Only patients who had pre-treatment CT scans were included in the study, and all included cases were confirmed to be locally advanced by clinical assessment and AJCC staging We excluded samples that were postoperative, had inferior image quality, or were devoid of sequencing data. The following conditions were defined as poor image quality: the presence of artifacts in the image (e.g., motion artifacts, metal artifacts caused by dentures) or missing/incomplete scanning of the tumor site. All data and images were anonymized and are publicly available, exempting them from ethical approval and informed consent. The TCIA dataset used in this study corresponds to the 'HNSCC collection,' which contains radiologic imaging data for patients with head and neck squamous cell carcinoma. Ethical approval was obtained from the relevant Institutional Review Board (IRB) or Ethics Committee. Since the data is publicly available and anonymized, informed consent was waived. All data used in this study complies with ethical guidelines for research. In this study, we concentrated on resampling to a consistent voxel size (1×1×1 mm³) across all images to minimize variability arising from differences in scanning equipment, protocols, and lesion sizes. While resampling helps standardize the images, it is important to note that other preprocessing methods, such as image discretization or intensity normalization, can further enhance the robustness and reproducibility of extracted features. Image discretization, which involves transforming continuous pixel values into discrete levels, could potentially improve the stability of radiomic features by reducing the impact of noise and inconsistencies in image intensity. However, prior research in radiomics has shown that resampling alone can often provide sufficient improvement in feature reproducibility when combined with careful segmentation. Given the high quality of the CT images from the TCIA dataset and the specific focus on arterial-phase images, we chose to focus primarily on resampling as our preprocessing step. Further exploration of additional preprocessing techniques, such as image discretization, may be considered in future studies to evaluate their impact on the robustness of radiomic features. All CT images were acquired using multi-detector computed tomography (MDCT) scanners from different institutions contributing to the TCIA database. The scanning protocols varied slightly depending on the institution; however, all images were contrast-enhanced arterial phase scans with slice thicknesses ranging from 1 to 3 mm. The tube voltage was typically set between 100–140 kVp, and the tube current ranged from 100–400 mA, adjusted automatically by the scanner’s dose modulation system. The field of view (FOV) and matrix size were standardized as much as possible to maintain consistency in image quality. Images were reconstructed using standard convolution kernels to optimize contrast resolution while minimizing artifacts. These standardized imaging parameters ensured the reproducibility and comparability of the radiomics analysis. Radiomic Features Extraction All images were resampled to a uniform voxel size of 1 × 1 × 1 mm³ to minimize variability caused by differences in scanning equipment, imaging protocols, and lesion size. Lesion segmentation was performed using 3D Slicer software (version 4.10.2; https://www.slicer.org/ ). Two experienced radiologists jointly identified lesion boundaries and delineated the volumes of interest (VOIs). Both radiologists were blinded to the clinical and pathological information throughout the segmentation process. Feature extraction was performed using Python’s open-source pyradiomics 3.0.1 package. A total of 1316 radiomic features were extracted from each CT image. These features comprised: First-order statistics features (n = 18): Describing the intensity distribution within the region of interest (ROI), including metrics such as mean, median, entropy, skewness, and kurtosis. Shape-based (3D) features (n = 14): Characterizing tumor geometry, such as volume, surface area, sphericity, elongation, and flatness. Texture features, including: Gray-Level Co-occurrence Matrix (GLCM) features (n = 24) Gray-Level Run Length Matrix (GLRLM) features (n = 16) Gray-Level Size Zone Matrix (GLSZM) features (n = 16) Gray-Level Dependence Matrix (GLDM) features (n = 14) Neighboring Gray Tone Difference Matrix (NGTDM) features (n = 5) Wavelet-transformed features (n = 1200): Derived by applying all possible high-pass (H) and low-pass (L) filter combinations to the original images, generating eight decomposed images per lesion. Construction and evaluation of radiomics models for TMB prediction After extraction, the TCIA dataset was randomly divided into training and validation datasets in a 7:3 ratio, resulting in 112 individuals in the training dataset and 47 in the validation dataset. The division was performed randomly and was not based on TMB status, prognosis, or any clinical characteristics. The training dataset was used for feature selection and model development, while the validation dataset was used to assess model performance. A signature selection strategy was applied, which included the following steps: univariate Cox regression analysis, least absolute shrinkage and selection operator (LASSO)-Cox regression with 5-fold cross-validation, and multivariate Cox regression analysis. The optimal Lambda Lasso_output_min consists of features for establishing a radiomics signature models. Afterwards, clinical models and radiomics signature models were constructed and visualized using nomogram. The performance of the ROC curve validation model was assessed to evaluate the discriminative ability of the radiomics signature in predicting TMB status. The same methodology was used for both TMB and OS prediction models; however, each model targets a distinct clinical outcome, TMB and OS, requiring independent analysis and validation for each. Construction and evaluation of radiomics models for OS prediction OS was defined as the duration from treatment initiation until any event resulting in patient death. If a patient remains alive at the last follow-up, the OS time is censored. Kaplan-Meier survival curves were used to estimate the OS rates, and both univariate and multivariate Cox proportional hazards regression analyses were performed to examine the determinants of OS. The same random 7:3 split used for the TMB prediction model was applied, with 112 individuals in the training dataset and 47 in the validation dataset. This division was not based on prognosis status or survival outcomes. Feature selection was performed using the training dataset, and model performance was evaluated on the validation dataset. We employed a signature selection strategy that consisted of the following steps: univariate Cox regression analysis, least absolute shrinkage and selection operator (LASSO)–Cox regression with 5-fold cross-validation, and multivariate Cox regression analysis. The optimal Lambda Lasso_output_min consists of features for establishing a radiomics signature models. Afterwards, clinical models and radiomics signature models were constructed and visualized using nomogram. The performance of the ROC curve validation model was assessed to evaluate the discriminative ability of the radiomics signature in predicting TMB status. Although identical methods were employed to select features and construct models for TMB and OS, the distinct nature of TMB (a biomarker of mutation burden) and OS necessitates separate presentation of results for each model. Clinical model construction The clinical data included age, sex, histological differentiation, T stage, N stage, M stage, survival time, and survival status. Univariate logistic regression analysis and multivariate analysis were performed on clinical data to construct a clinical model for predicting TMB. Additionally, univariate and multivariate Cox proportional hazards regression analyses were conducted to identify factors associated with OS. For the TMB clinical model, univariate logistic regression analysis was used to assess the association between clinical variables and TMB status (high vs. low). Variables with a p-value < 0.1 in the univariate analysis were entered into the multivariate logistic regression model to identify independent predictors and construct the final model. For the OS clinical model, univariate Cox proportional hazards regression analysis was performed to evaluate the prognostic significance of clinical variables for OS. Variables with a p-value < 0.1 in the univariate analysis were included in the multivariate Cox regression to identify independent prognostic factors and develop the final model. Statistical Analysis Statistical analysis and model development were performed using R 3.3.3 software (IBM, Armonk, NY, USA). The standardized variables were expressed as mean ± standard deviation (SD) or as median (25th percentile, 75th percentile), and evaluated using univariate logistic regression analysis, univariate Cox regression analysis, and multivariate Cox regression analysis. OS was defined as the duration from randomization to the occurrence of any event resulting in patient death. OS rates for patients in both groups were assessed through Kaplan-Meier analysis. Both univariate and multivariate Cox proportional hazards regression analyses were performed to identify the determinants of OS. P < 0.05 was considered statistically significant. The DeLong test was used to compare the AUC values between the training and validation datasets. ROC curves were plotted using the "pROC" package. Calibration curves were generated using the "rms" package, and decision curve analysis (DCA) was performed using the "dca" function.” For the TMB prediction task, classification methods (logistic regression and LASSO-based feature selection) were employed, while OS prediction used Cox proportional hazards regression appropriate for time-to-event data. Results 1 Mutations in HNSCC patients The TCGA-HNSCC dataset includes data from 506 patients with head and neck squamous cell carcinoma. DEGs were identified using |log₂FC| >1.0 and an FDR < 0.05 as the cut-off criteria. A total of 136 genes were upregulated and 293 genes were downregulated (Fig. 1). To further investigate the biological significance of these DEGs, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were performed. GO enrichment analysis revealed that the DEGs were mainly involved in immune system processes, signaling receptor activity, and plasma membrane development (Fig. 2A–C). KEGG enrichment analysis indicated that the DEGs were primarily enriched in cytokine–cytokine receptor interaction, focal adhesion, and cell cycle pathways (Fig. 2D). 2 Analysis of TMB and clinical relevance in HNSCC The cut-off value for TMB was determined based on the AUC values for survival time using X-tile software. Patients were subsequently divided into low- and high-TMB groups, with a threshold of 4.2 mutations/Mb (Table 1). Analysis of somatic mutations revealed that the top 10 most frequently mutated genes in HNSCC were TP53 (70.3%), TTN (40.2%), FAT1 (23.4%), CDKN2A (20.8%), MUC16 (19.2%), CSMD3 (18.5%), PIK3CA (17.7%), NOTCH1 (17.5%), SYNE1 (17.1%), and LRP1B (16.4%) (Fig. 3A). Kaplan-Meier survival analysis demonstrated that patients in the high-TMB group had significantly worse OS compared to those in the low-TMB group (HR = 1.65, 95% CI: 1.20–2.28, p = 0.002; Fig. 3B). 3 Construction and evaluation of radiomics models for Prediction TMB A total of 159 HNSCC patients from the TCIA database were included in this study. To ensure unbiased validation, the dataset was randomly divided into a training cohort (n = 112) and a validation cohort (n = 47) at a 7:3 ratio. The training and validation cohorts were comparable in terms of clinical variables, molecular subtypes, and TMB status. Univariate COX regression analysis identified a total of 1316 features with statistically significant differences between the Low-TMB and High-TMB groups in the training set. Key features selected for the TMB prediction model included texture-based features, such as Gray-Level Co-occurrence Matrix (GLCM) entropy, which reflects the heterogeneity in the tumor's intensity patterns, and volume-based shape features, involving tumor volume and surface area, which are critical indicators of tumor growth and aggressiveness. These features were associated with higher TMB levels, as increased tumor heterogeneity often correlates with a greater mutation burden. Using LASSO regression, the 1316 features were narrowed down to 16 key features. Notably, wavelet-transformed texture features, such as GLCM contrast and GLDM dependence variance, were selected as significant predictors. These features highlight the complexity and variation in tumor texture, which may reflect underlying biological differences in TMB, such as higher genomic instability and mutation rates (Fig. 4A, B). In the radiomics model for TMB prediction, key features, such as GLCM entropy, surface area-to-volume ratio, and wavelet-transformed contrast emerged as the most important predictors. Tumors with high entropy and contrast values tend to show more disordered texture and heterogeneity, characteristics that have been linked to higher mutation burden. These findings support the hypothesis that complex tumor textures are associated with greater genomic instability. The radiomics model showed strong discrimination in the training set (AUC = 0.881); however, its validation set performance (AUC = 0.64) indicates moderate predictive ability. When combined with clinical features, the model’s performance improved (AUC = 0.902 in training and 0.669 in validation), suggesting added value in integrating radiomic and clinical features (Fig. 4C, D). Univariate and multivariate logistic regression analyses revealed that TMB status was significantly associated only with the primary tumor site (p < 0.001). Among different tumor sites, laryngeal cancer exhibited the highest TMB values, while oropharyngeal cancer showed the lowest. The clinical model achieved AUC values of 0.596 and 0.575 in the training and validation cohorts, respectively. In contrast, the combined model demonstrated improved predictive performance for TMB status, with AUC values of 0.902 in the training cohort and 0.669 in the validation cohort (Fig. 4C, D). 5 Construction and evaluation of radiomics models for Prediction OS In addition, radiomics models were utilized to predict OS in HNSCC patients. A total of 159 HNSCC patients from the TCIA database were included in this study. Based on the grouping used for TMB prediction models, feature selection was performed in the training dataset, which consisted of 112 patients, while 47 patients were allocated to the validation dataset. Univariate COX regression analysis identified key features associated with survival outcomes, including first-order features, such as mean intensity and skewness, which capture the overall distribution of voxel intensities within the tumor. Higher mean intensity and more positive skewness were correlated with worse prognosis, reflecting the presence of areas with more aggressive tumor growth. Additionally, shape features, including tumor sphericity and elongation were also important, as tumors with irregular shapes tend to exhibit more invasive behavior and poorer outcomes. The LASSO logistic regression model, with five-fold cross-validation under the minimum criterion, was applied to reduce the 1,316 features to 10 selected features and to construct the radiomics signature (Fig. 5A-B). The radiomics model for OS prediction identified first-order features such as mean intensity and skewness, along with shape features like sphericity. Higher mean intensity and less spherical tumors were associated with worse survival, suggesting that more irregular tumors may be more aggressive and less responsive to treatment. These results underscore the importance of tumor texture and shape in predicting long-term patient outcomes. The model’s AUC values in the training and validation sets were 0.817, 0.672, and 0.601, and 0.729, 0.715, and 0.711, respectively (Fig. 5C, D). Univariate and multivariate logistic regression analysis revealed that T stage and gender were significantly associated with worse OS (OS) (p < 0.05). The clinical model’s AUC values for the 1-, 3-, and 5-year survival rates in the training and validation sets were 0.626, 0.677, and 0.651, and 0.667, 0.607, and 0.583, respectively (Fig. 5E, F). 6 Using Combined model to Prognosticate OS Variables that were significant in both univariable and multivariable analyses (T stage, gender, and radiomics features) were used to develop the combined model nomogram (Fig. 6A). After evaluation, the combined model demonstrated favorable predictive ability for OS, as indicated by the ROC curve for 1-, 3-, and 5-year survival rates. The model’s AUC values in the training and validation sets were 0.784, 0.693, and 0.665, and 0.772, 0.734, and 0.680, respectively (Fig. 6B, C). The Kaplan–Meier survival curves for the combined model are shown in Figs. 6D and E. Furthermore, a high-risk combined model was significantly associated with lower OS in both the training and validation sets (p < 0.05). Discussion HNSCC is the sixth most common malignancy worldwide [ 1 ]. Despite significant advances in screening, diagnosis, and treatment, particularly with the advent of immunotherapy, HNSCC remains associated with poor prognosis [ 4 ]. Increasingly, cancer genomic data, including somatic mutations, copy number variations, and TMB, are being integrated to improve prognosis prediction and treatment response. A comprehensive analysis of 45 clinical studies involving 103,078 cancer patients suggested that high TMB may be an unfavorable prognostic indicator for patients receiving non-immunotherapy treatments. However, in patients treated with immunotherapy, high TMB has been globally associated with improved survival and better treatment efficacy [ 16 ]. In our study, we examined the prognostic significance of TMB and its correlation with immune cell infiltration in HNSCC, using the TCGA dataset. Our findings indicate that high-TMB status correlates with poor OS, which aligns with a previous multicenter retrospective study in Germany [ 17 ]. In recent years, ICIs have shown promising therapeutic potential in recurrent/metastatic (R/M) HNSCC. Pembrolizumab and nivolumab, monoclonal antibodies targeting the PD-1 receptor, have demonstrated durable therapeutic efficacy in HNSCC [ 18 , 19 ]. Studies have also indicated that TMB, particularly defined as the total number of somatic mutations per coding region of a tumor genome, can be a predictor of response to ICIs in R/M HNSCC [ 20 – 24 ]. Patients with high TMB in the KEYNOTE-012 and KEYNOTE-055 trials showed efficacy in response to pembrolizumab, defined as ≥ 10 mut/Mb [ 25 ]. TMB has emerged as a potential biomarker for predicting immunotherapy response in prospective clinical trials of HNSCC. However, the testing methods for TMB, such as next-generation sequencing (NGS) and whole exome sequencing (WES), are costly, time-consuming, and require fresh samples. The need for a more accessible, cost-effective alternative is evident. In this regard, CT radiomics has surfaced as a promising approach to identify potential immunotherapy responders and improve patient stratification for treatment decisions [ 28 ]. With the rapid advancement of artificial intelligence (AI) and machine learning, these methods now offer substantial potential for integrating multi-omics datasets to improve clinical decision-making in HNSCC treatment [ 29 ]. Previous radiomics studies in head and neck tumors have shown the ability of radiomics models to predict various tumor characteristics, such as TMB and immunotherapy responses. For instance, CT-based radiomics models have successfully predicted TMB in cancers like non-small cell lung cancer and lower-grade gliomas [ 35 – 37 , 38 ]. However, no studies to date have used radiomics to predict TMB in HNSCC patients. In our study, a CT-based radiomics model was developed to predict TMB, demonstrating good performance in both the training and validation sets (AUC = 0.881 and 0.64, respectively). These findings suggest that radiomics could be a valuable tool for TMB expression detection, particularly when biopsy is not feasible. While our model showed promising results, it is important to note that this radiomics signature model may complement, rather than replace, traditional methods for TMB detection, such as tissue biopsy and sequencing. The predictive ability of our model could also complement the TNM staging system. This model provides supplementary information that could influence treatment decisions, particularly in determining which patients may benefit from treatment deintensification. However, it is essential to consider that our study does not directly address how this model could change clinical decision-making or guide treatment strategies. While our radiomics-based approach may assist in identifying patients who are likely to benefit from immunotherapy, further investigation is required to explore its role in treatment personalization and clinical practice. There are some limitations to this study. Firstly, the use of data from public databases may introduce variability and bias, especially in the validation process. External cohort validation is essential to confirm the applicability of this model across different institutions and imaging systems. Secondly, manual lesion delineation introduces human error and variability, which could affect the reproducibility of the results. Future research should focus on developing automated, robust segmentation techniques to reduce these biases. Thirdly, overfitting remains a concern, particularly given the discrepancy in AUC values between the training (0.881) and validation sets (0.64). While the model shows good predictive ability, the lower performance in the validation set suggests the need for further refinement. Additionally, the comparison of radiomics-based approaches with other TMB prediction methods, such as circulating tumor DNA (ctDNA) or machine learning models using clinical features, was not explored in this study. Although the differences in clinical characteristics and prognosis between high- and low-TMB groups were analyzed using TCGA data, a similar analysis could not be performed in the TCIA dataset due to the limited availability of clinical and follow-up data. Instead, the TCIA data were utilized primarily for the development and validation of radiomics-based models for TMB and OS prediction. Despite the moderate AUC values in validation, the combined model showed a significant association with OS, suggesting its potential utility in clinical decision-making with further refinement. Future research should address this gap by comparing the effectiveness of these methods in predicting TMB and their potential to guide treatment strategies. To improve the generalizability of our findings, future studies will incorporate independent external validation cohorts and assess the models' performance across diverse populations. Further optimization of feature selection methods and model training strategies, such as deep learning approaches or ensemble models, may enhance predictive accuracy. Conclusion In conclusion, TMB status serves as a significant and independent prognostic factor for HNSCC. The combined model, based on enhanced CT images, demonstrates potential in predicting TMB status and shows promise for prognostic assessment in HNSCC. Leveraging this model may aid clinicians in making more informed decisions and support individualized and precise diagnosis and treatment strategies for HNSCC patients. However, further validation, particularly with external datasets, is necessary to confirm its generalizability and clinical utility. Declarations Ethics approval and consent to participate Not applicable. Patient consent for publication Not applicable. Data availability The data that support the fndings of this study are available from TCIA (https://www.cancerimagingarchive. net) and TCGA (https://portal.gdc.cancer.gov). Funding This work was supported by University of Electronic Science and Technology, Sichuan Cancer Hospital Oncology Medical Industry Innovation Fund (ZYGX2021 YGCX006) , Sichuan Science and Technology Department Key Research and Development Project 2023YFS0157). Authors' contributions YKL and NH designed the study. LL and YJQ conducted data analyses and drafted the manuscript. MF, YQ, JRX, and JLR contributed to the data analyses. All authors participated in the manuscript's development and approved the final version. This research was supported by the Fundamental Research Funds for the Central Universities (ZYGX2021YGCX006). References Bray, F., Ferlay, J., Soerjomataram, I., et al. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2018;68(6):394–424. https://doi.org/10.3322/caac.21492. NCCN Clinical Practice Guidelines in Oncology (NCCN Guidelines®). Head and Neck Cancers Version 2.2023. https://www.nccn.org/professionals/physician_gls/pdf/head-and-neck.pdf. Burtness, B., Harrington, K.J., Greil, R., et al. Pembrolizumab alone or with chemotherapy versus cetuximab with chemotherapy for recurrent or metastatic squamous cell carcinoma of the head and neck (KEYNOTE-048): a randomised, open-label, phase 3 study. Lancet. 2019;394(10212):1915–1928. https://doi.org/10.1016/S0140-6736(19)32591-7. Cohen, E.E., Ferris, R.L., Psyrri, A., et al. Primary results of the phase III JAVELIN head & neck 100 trial: avelumab plus chemoradiotherapy (CRT) followed by avelumab maintenance vs CRT in patients with locally advanced squamous cell carcinoma of the head and neck (LA SCCHN). Ann Oncol. 2020;31(Suppl 4):S658 . https://doi.org/10.1016/j.annonc.2020.08.1016. Lee, N.Y., Ferris, R.L., Psyrri, A., et al. Avelumab plus standard-of-care chemoradiotherapy versus chemoradiotherapy alone in patients with locally advanced squamous cell carcinoma of the head and neck: a randomised, double-blind, placebo-controlled, multicentre, phase 3 trial. Lancet Oncol. 2021;22(4):450–462. https://doi.org/10.1016/S1470-2045(21)00076-0. Machiels, J., Tao, Y., Burtness, B., et al. Primary results of the phase III KEYNOTE-412 study: pembrolizumab (pembro) with chemoradiation therapy (CRT) vs placebo plus CRT for locally advanced (LA) head and neck squamous cell carcinoma (HNSCC). Ann Oncol. 2022;33(suppl_7):S808 –S869. https://doi.org/10.1016/j.annonc.2022.07.714. Yilmaz, E., Ismaila, N., Dabney, R., et al. Immunotherapy and Biomarker Testing in Recurrent and Metastatic Head and Neck Cancers: ASCO Guideline Q and A. JCO Oncol Pract. 2023;19(4):194–196. https://doi.org/10.1200/OP.22.00789. Litchfield, K., Reading, J.L., Puttick, C., et al. Meta-analysis of tumor- and T cell-intrinsic mechanisms of sensitization to checkpoint inhibition. Cell. 2021;184(3):596–614.e14. https://doi.org/10.1016/j.cell.2021.01.002. Bera, K., Braman, N., Gupta, A., et al. Predicting cancer outcomes with radiomics and artificial intelligence in radiology. Nat Rev Clin Oncol. 2022;19(2):132–146. https://doi.org/10.1038/s41571-021-00560-7. Mayerhoefer, M.E., Materka, A., Langs, G., et al. Introduction to Radiomics. J Nucl Med. 2020;61(4):488–495. https://doi.org/10.2967/jnumed.119.232231. Zhu, Y., Wu, M. Noninvasive radiomic analysis of enhanced CT predicts CTLA4 expression and prognosis in head and neck squamous cell carcinoma. Sci Rep. 2023;13(1):16782. https://doi.org/10.1038/s41598-023-44055-0. Ye, J., Luo, J., Xu, S., et al. One-slice CT image based kernelized radiomics model for the prediction of low/mid-grade and high-grade HNSCC. Comput Med Imaging Graph. 2020;80:101675. https://doi.org/10.1016/j.compmedimag.2020.101675. Bruixola, G., et al. Radiomics and radiogenomics in head and neck squamous cell carcinoma: Potential contribution to patient management and challenges. Cancer Treat Rev. 2021;99:102263. https://doi.org/10.1016/j.ctrv.2021.102263. Yu, G., Wang, L.G., Han, Y., He, Q.Y. clusterProfiler: an R package for comparing biological themes among gene clusters. OMICS. 2012;16(5):284–287. https://doi.org/10.1089/omi.2011.0118. Lv, J., Zhu, Y., Ji, A., Zhang, Q., Liao, G. Mining TCGA database for tumor mutation burden and their clinical significance in bladder cancer. Biosci Rep. 2020;40(4):BSR20194337 . https://doi.org/10.1042/BSR20194337. Cao, D., Xu, H., Xu, X., Guo, T., Ge, W. High tumor mutation burden predicts better efficacy of immunotherapy: a pooled analysis of 103078 cancer patients. Oncoimmunology. 2019;8(9):e1629258 . https://doi.org/10.1080/2162402X.2019.1629258. Eder, T., Hess, A.K., Konschak, R., et al. Interference of tumour mutational burden with outcome of patients with head and neck cancer treated with definitive chemoradiation: a multicentre retrospective study of the German Cancer Consortium Radiation Oncology Group. Eur J Cancer. 2019;116:67–76. https://doi.org/10.1016/j.ejca.2019.04.015. Burtness, B., et al. Pembrolizumab alone or with chemotherapy versus cetuximab with chemotherapy for recurrent or metastatic squamous cell carcinoma of the head and neck (KEYNOTE-048): a randomised, open-label, phase 3 study. Lancet. 2019;394(10212):1915–1928. https://doi.org/10.1016/S0140-6736(19)32591-7. Ferris, R.L., et al. Nivolumab vs investigator's choice in recurrent or metastatic squamous cell carcinoma of the head and neck: 2-year long-term survival update of CheckMate 141 with analyses by tumor PD-L1 expression. Oral Oncol. 2018;81:45–51. https://doi.org/10.1016/j.oraloncology.2018.04.008. Zhang, Y., Lin, A., Li, Y., et al. Age and mutations as predictors of the response to immunotherapy in head and neck squamous cell cancer. Front Cell Dev Biol. 2020;8:608969. https://doi.org/10.3389/fcell.2020.608969. Wildsmith, S., Li, W., Wu, S., et al. Tumor Mutational Burden as a Predictor of Survival with Durvalumab and/or Tremelimumab Treatment in Recurrent or Metastatic Head and Neck Squamous Cell Carcinoma. Clin Cancer Res. 2023;29(11):2066–2074. https://doi.org/10.1158/1078-0432.CCR-22-2765. Chen, S., Yang, Y., He, S., et al. Review of biomarkers for response to immunotherapy in HNSCC microenvironment. Front Oncol. 2023;13:1037884. https://doi.org/10.3389/fonc.2023.1037884. Valero, C., Golkaram, M., Vos, J.L., et al. Clinical-genomic determinants of immune checkpoint blockade response in head and neck squamous cell carcinoma. J Clin Invest. 2023;133(19):e169339 . https://doi.org/10.1172/JCI169339. Haddad, R.I., Seiwert, T.Y., Chow, L.Q.M., et al. Influence of tumor mutational burden, inflammatory gene expression profile, and PD-L1 expression on response to pembrolizumab in head and neck squamous cell carcinoma. J Immunother Cancer. 2022;10(2):e003940 . https://doi.org/10.1136/jitc-2021-003940. Pfister, D.G., Haddad, R.I., Worden, F.P., et al. Biomarkers predictive of response to pembrolizumab in head and neck cancer. Cancer Med. 2023;12(6):6603–6614. https://doi.org/10.1002/cam4.5478. Rodrigo, J.P., Sánchez-Canteli, M., Otero-Rosales, M., et al. Tumor mutational burden predictability in head and neck squamous cell carcinoma patients treated with immunotherapy: systematic review and meta-analysis. J Transl Med. 2024;22(1):135. https://doi.org/10.1186/s12967-024-04935-z. Verdegaal, E.M., de Miranda, N.F., Visser, M., et al. Neoantigen landscape dynamics during human melanoma-T cell interactions. Nature. 2016;536(7614):91–95. https://doi.org/10.1038/nature18945. Rodrigo, J.P., et al. Tumor mutational burden predictability in head and neck squamous cell carcinoma patients treated with immunotherapy: systematic review and meta-analysis. J Transl Med. 2024;22(1):135. https://doi.org/10.1186/s12967-024-04935-z. Van den Bossche, V., Zaryouh, H., Vara-Messler, M., et al. Microenvironment-driven intratumoral heterogeneity in head and neck cancers: clinical challenges and opportunities for precision medicine. Drug Resist Updat. 2022;60:100806. https://doi.org/10.1016/j.drup.2022.100806. Dong, C., Zheng, Y.M., Li, J., et al. A CT-based radiomics nomogram for differentiation of squamous cell carcinoma and non-Hodgkin's lymphoma of the palatine tonsil. Eur Radiol. 2022;32(1):243–253. https://doi.org/10.1007/s00330-021-08143-x. Zheng, Y.M., Xu, W.J., Hao, D.P., et al. A CT-based radiomics nomogram for differentiation of lympho-associated benign and malignant lesions of the parotid gland. Eur Radiol. 2021;31(5):2886–2895. https://doi.org/10.1007/s00330-020-07388-2. Zheng, Y.M., Li, J., Liu, S., et al. MRI-Based radiomics nomogram for differentiation of benign and malignant lesions of the parotid gland. Eur Radiol. 2021;31(6):4042–4052. https://doi.org/10.1007/s00330-021-07796-y. Zhao, L., Gong, J., Xi, Y., et al. MRI-based radiomics nomogram may predict the response to induction chemotherapy and survival in locally advanced nasopharyngeal carcinoma. Eur Radiol. 2020;30(1):537–546. https://doi.org/10.1007/s00330-019-06371-w. Gillies, R.J., Kinahan, P.E., Hricak, H. Radiomics: images are more than pictures, they are data. Radiology. 2016;278(2):563–577. https://doi.org/10.1148/radiol.2015151169. Wang, J., Wang, J., Huang, X., et al. CT radiomics-based model for predicting TMB and immunotherapy response in non-small cell lung cancer. BMC Med Imaging. 2024;24(1):45. https://doi.org/10.1186/s12880-024-01216-5. Wen, Q., Yang, Z., Dai, H., et al. Radiomics Study for Predicting the Expression of PD-L1 and Tumor Mutation Burden in Non-Small Cell Lung Cancer Based on CT Images and Clinicopathological Features. Front Oncol. 2021;11:620246. https://doi.org/10.3389/fonc.2021.620246. Yang, J., Shi, W., Yang, Z., et al. Establishing a predictive model for tumor mutation burden status based on CT radiomics and clinical features of non-small cell lung cancer patients. Transl Lung Cancer Res. 2023;12(4):808–823. https://doi.org/10.21037/tlcr-23-123. LHT, Chu, N.T., Tran, T.O., et al. A Radiomics-Based Machine Learning Model for Prediction of Tumor Mutational Burden in Lower-Grade Gliomas. Cancers (Basel). 2022;14(14):3492. https://doi.org/10.3390/cancers14143492. Ma, T., Zhang, Y., Zhao, M., et al. A machine learning-based radiomics model for prediction of tumor mutation burden in gastric cancer. Front Genet. 2023;14:1283090. https://doi.org/10.3389/fgene.2023.1283090. Lee, H., Moon, S.H., Hong, J.Y., et al. A Machine Learning Approach Using FDG PET-Based Radiomics for Prediction of Tumor Mutational Burden and Prognosis in Stage IV Colorectal Cancer. Cancers (Basel). 2023;15(15):3841. https://doi.org/10.3390/cancers15153841. Zheng, Y., Yuan, M., Zhou, R., et al. A computed tomography-based radiomics signature for predicting expression of programmed death ligand 1 in head and neck squamous cell carcinoma. Eur Radiol. 2022;32(8):5362–5370. https://doi.org/10.1007/s00330-022-08618-5. Yoon, J., Suh, Y.J., Han, K., et al. Utility of CT radiomics for prediction of PD-L1 expression in advanced lung adenocarcinomas. Thorac Cancer. 2020;11(4):993–1004. https://doi.org/10.1111/1759-7714.13352. Patel, H., Vock, D.M., Marai, G.E., et al. Oropharyngeal cancer patient stratification using random forest based learning over high-dimensional radiomic features. Sci Rep. 2021;11:1–11. https://doi.org/10.1038/s41598-021-84444-x. Tang, F.H., Chu, C.Y.W., Cheung, E.Y.W. Radiomics AI prediction for head and neck squamous cell carcinoma (HNSCC) prognosis and recurrence with target volume approach. BJR Open. 2021;3(1):20200073. https://doi.org/10.1259/bjro.20200073. Corino, V.D.A., Bologna, M., Calareso, G., et al. A CT-Based Radiomic Signature Can Be Prognostic for 10-Months Overall Survival in Metastatic Tumors Treated with Nivolumab: An Exploratory Study. Diagnostics (Basel). 2021;11(6):1006. https://doi.org/10.3390/diagnostics11061006. Han, K., Joung, J.F., Han, M., et al. Locoregional Recurrence Prediction Using a Deep Neural Network of Radiological and Radiotherapy Images. J Pers Med. 2022;12(2):256. https://doi.org/10.3390/jpm12020256. Bernatz, S., Böth, I., Ackermann, J., et al. Radiomics for therapy-specific head and neck squamous cell carcinoma survival prognostication (part I). BMC Med Imaging. 2023;23(1):71. https://doi.org/10.1186/s12880-023-01026-1. Mes, S.W., van Velden, F.H.P., Peltenburg, B., et al. Outcome prediction of head and neck squamous cell carcinoma by MRI radiomic signatures. Eur Radiol. 2020;30(11):6311–6321. https://doi.org/10.1007/s00330-020-06971-x. Table 1 Table 1 is available in the Supplementary Files section. Additional Declarations No competing interests reported. Supplementary Files Table1.docx Cite Share Download PDF Status: Posted Version 1 posted 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-6921523","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":486006132,"identity":"921982ed-7d50-4f29-9e1a-4cbc423f3f9c","order_by":0,"name":"Lu Li","email":"","orcid":"","institution":"Sichuan Cancer Hospital \u0026 Institute, Sichuan Cancer Center, Affiliated Cancer Hospital of University of Electronic Science and Technology of China","correspondingAuthor":false,"prefix":"","firstName":"Lu","middleName":"","lastName":"Li","suffix":""},{"id":486006135,"identity":"cd1ccbf5-1cb1-41bb-9a68-d90b37a251b6","order_by":1,"name":"Youjing Qiu","email":"","orcid":"","institution":"Jinjiang Daguan Hospital","correspondingAuthor":false,"prefix":"","firstName":"Youjing","middleName":"","lastName":"Qiu","suffix":""},{"id":486006136,"identity":"0fbf79ca-31ea-4f78-87a7-e74aa367e855","order_by":2,"name":"Yuan Qin","email":"","orcid":"","institution":"Sichuan Cancer Hospital \u0026 Institute, Sichuan Cancer Center, Affiliated Cancer Hospital of University of Electronic Science and Technology of China","correspondingAuthor":false,"prefix":"","firstName":"Yuan","middleName":"","lastName":"Qin","suffix":""},{"id":486006137,"identity":"97a74c1d-4189-4e76-883b-7aee31584192","order_by":3,"name":"Jirui Xie","email":"","orcid":"","institution":"Sichuan Cancer Hospital \u0026 Institute, Sichuan Cancer Center, Affiliated Cancer Hospital of University of Electronic Science and Technology of China","correspondingAuthor":false,"prefix":"","firstName":"Jirui","middleName":"","lastName":"Xie","suffix":""},{"id":486006138,"identity":"645a6c4e-2db1-431d-9054-b6a9b18d3893","order_by":4,"name":"Jianlan Ren","email":"","orcid":"","institution":"Sichuan Cancer Hospital \u0026 Institute, Sichuan Cancer Center, Affiliated Cancer Hospital of University of Electronic Science and Technology of China","correspondingAuthor":false,"prefix":"","firstName":"Jianlan","middleName":"","lastName":"Ren","suffix":""},{"id":486006140,"identity":"ea1f555f-35cd-4417-91ec-4052d904a3cd","order_by":5,"name":"Mei Feng","email":"","orcid":"","institution":"Sichuan Cancer Hospital \u0026 Institute, Sichuan Cancer Center, Affiliated Cancer Hospital of University of Electronic Science and Technology of China","correspondingAuthor":false,"prefix":"","firstName":"Mei","middleName":"","lastName":"Feng","suffix":""},{"id":486006142,"identity":"fec1e496-cd46-42a7-b467-30e8270b912a","order_by":6,"name":"Na Huang","email":"","orcid":"","institution":"Sichuan Cancer Hospital \u0026 Institute, Sichuan Cancer Center, Affiliated Cancer Hospital of University of Electronic Science and Technology of China","correspondingAuthor":false,"prefix":"","firstName":"Na","middleName":"","lastName":"Huang","suffix":""},{"id":486006145,"identity":"d4042d26-9c9e-44ba-bddc-cbf2725d83d4","order_by":7,"name":"Yangkun Luo","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA70lEQVRIiWNgGAWjYBACPiBmbKhg4GGQIFYLG1jLGZK1NLYBSeK1SCQfk5w5r06Gf3aP4eeCGrs8/vbeBwwf99Ti0ZKWJrlx22EeiTtnjKVnHEsuljhz3IBxxrPjeLTkmEk+3HaAx0Aix0Cat4E5cYNEGgMzz4FjBLTMqQNpMf7N21CfuEH+GRFaNjYwg7SYAW05DLSFDaSlBrcWnmfJljOOAf1yI63MmufY8cQZZ9IYDs44cACnFn725IM3e2rq7PlnJG++zVNTndjffozxwYcDdTi1AAELZowArTiMTwvzB2yieG0ZBaNgFIyCkQUArAFPOfYjSOsAAAAASUVORK5CYII=","orcid":"","institution":"Sichuan Cancer Hospital \u0026 Institute, Sichuan Cancer Center, Affiliated Cancer Hospital of University of Electronic Science and Technology of China","correspondingAuthor":true,"prefix":"","firstName":"Yangkun","middleName":"","lastName":"Luo","suffix":""}],"badges":[],"createdAt":"2025-06-18 09:38:11","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6921523/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6921523/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":87319260,"identity":"eea5c0de-dc33-4534-959a-1fa2e15a9f15","added_by":"auto","created_at":"2025-07-22 16:21:15","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":421980,"visible":true,"origin":"","legend":"\u003cp\u003eHeatmap showing the top 40 DEGs\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-6921523/v1/1eb91fc232e3406a3541bf4d.png"},{"id":87319262,"identity":"3a7045d8-2fce-4053-b4bb-8d9b2affe4f0","added_by":"auto","created_at":"2025-07-22 16:21:15","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":192790,"visible":true,"origin":"","legend":"\u003cp\u003eA-C: Functional analysis of the top 10 enriched biological processes (BPs), cell composition (CC), and molecular function (MF) of GO analysis); D: KEGG enrichment diseases analysis\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-6921523/v1/01233af0fb79f1afd58cd369.png"},{"id":87319270,"identity":"8995f0cd-cb70-4a2d-996c-5820ae067e77","added_by":"auto","created_at":"2025-07-22 16:21:15","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":385511,"visible":true,"origin":"","legend":"\u003cp\u003eA: Waterfall plot of the top 30 mutated genes in the TCGA HNSCC cohort; B: Kaplan-Meier curves of overall survival of the high- and low-TMB groups.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-6921523/v1/0fa0b48b7db2245a2012bea5.png"},{"id":87319269,"identity":"ef176647-cad6-4dfa-9784-dbc3dcb53c81","added_by":"auto","created_at":"2025-07-22 16:21:15","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":155540,"visible":true,"origin":"","legend":"\u003cp\u003eA-B: Radiomics feature selection using the least absolute shrinkage and selection operator (LASSO) regression model; C-D: Receiver operating characteristics curves of the radiomics signature, clinical model, and combined model in the training and validation datasets.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-6921523/v1/ee0b3257203d5896814bdb97.png"},{"id":87319266,"identity":"297bb5cd-4caa-4c07-9fe3-f74bad72faf9","added_by":"auto","created_at":"2025-07-22 16:21:15","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":239234,"visible":true,"origin":"","legend":"\u003cp\u003eA-B: Radiomics feature selection using the least absolute shrinkage and selection operator (LASSO) regression model; C-D: Receiver operating characteristics curves of the radiomics signature in the training and validation datasets; E-F: Receiver operating characteristics curves of the clinical model in the training and validation datasets.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-6921523/v1/a54b553b0c587442467ec4de.png"},{"id":87319264,"identity":"199001bb-3df4-4bc4-bbc4-22a0564b8b14","added_by":"auto","created_at":"2025-07-22 16:21:15","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":234679,"visible":true,"origin":"","legend":"\u003cp\u003eA: The constructed combined model nomogram；B-C: Receiver operating characteristics curves of the combined model in the training and validation datasets; D-E: Kaplan-Meier curves for prognostication using combined model.\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-6921523/v1/33ec08a26e0c5b0563c8c902.png"},{"id":88867342,"identity":"81d7e7f5-363c-42d2-b8d8-60b48f3e476f","added_by":"auto","created_at":"2025-08-12 08:39:02","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2411868,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6921523/v1/8108df85-3b18-4150-b732-65d0cfdfeb9c.pdf"},{"id":87319257,"identity":"4c250a57-1098-42bf-94dd-9c1e8e46b6d1","added_by":"auto","created_at":"2025-07-22 16:21:15","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":17789,"visible":true,"origin":"","legend":"","description":"","filename":"Table1.docx","url":"https://assets-eu.researchsquare.com/files/rs-6921523/v1/14d7e4b7b0996b50a367e54b.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"CT Radiomics-Clinical Model for Noninvasive Prediction of Tumor Mutation Burden and Survival in Locally Advanced Head and Neck Squamous Cell Carcinoma","fulltext":[{"header":"Introduction","content":"\u003cp\u003eHead and neck squamous cell carcinoma (HNSCC) is a major cause of cancer-related mortality, with locally advanced (LA) stages presenting significant treatment challenges and poor prognosis. Despite the use of chemoradiotherapy as the standard treatment for LA-HNSCC, outcomes remain suboptimal, with a high rate of recurrence and metastasis. Recent studies, such as the JAVELIN Head and Neck 100 trial and the KEYNOTE-412 study, explored the efficacy of combining PD-L1 or PD-1 inhibitors with chemoradiotherapy but failed to show significant improvements in overall survival (OS). The KEYNOTE-412 trial indicated a trend toward improved progression-free survival (PFS) for patients with a high combined positive score (CPS\u0026thinsp;≧\u0026thinsp;20), yet OS benefits remained limited. These findings highlight the need for more effective strategies to predict treatment response and improve survival outcomes for LA-HNSCC patients.\u003c/p\u003e\u003cp\u003eNumerous studies have explored biomarkers predictive of response to immune checkpoint inhibitors (ICIs), including PD-L1 expression, Combined Positive Score (CPS), and tumor mutational burden (TMB) [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Among these, a pan-cancer analysis identified TMB\u0026mdash;particularly clonal TMB\u0026mdash;as a robust predictor of response to ICI therapy [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Whole-exome sequencing (WES) remains the gold standard for TMB measurement; however, its widespread clinical implementation is hindered by high costs, extended turnaround times, and limited accessibility.\u003c/p\u003e\u003cp\u003eRadiomics, a high-throughput image analysis technique, enables the extraction of a large number of quantitative features from standard medical images, capturing intratumoral heterogeneity that is invisible to the naked eye. By converting biomedical images into mineable data, radiomics has shown promise in preoperative evaluation, tumor classification, prognosis assessment, and treatment response prediction [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Compared to WES, radiomics offers a noninvasive, cost-effective, and widely accessible alternative for evaluating tumor biology in clinical practice.\u003c/p\u003e\u003cp\u003eRadiogenomics, the integration of radiomics and genomic data, provides a framework to infer genomic and molecular characteristics\u0026mdash;such as gene expression profiles, methylation patterns, and TMB\u0026mdash;from imaging features. This approach enhances clinical decision-making by linking radiological phenotypes with underlying tumor biology [\u003cspan additionalcitationids=\"CR12\" citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. Importantly, radiomics-based TMB prediction may offer a clinically feasible and scalable method for immunotherapy stratification in settings where genomic testing is impractical.\u003c/p\u003e\u003cp\u003eThe present study aimed to develop a noninvasive radiomics approach to predict TMB status and survival outcomes in patients with LA-HNSCC. Genomic data were retrieved from The Cancer Genome Atlas (TCGA) to calculate TMB and analyze its associations with clinical characteristics, immune cell infiltration, and related gene expression. Radiomic features were extracted from computed tomography (CT) images of 159 patients obtained from The Cancer Imaging Archive (TCIA). Using these features, we constructed predictive models for TMB status and evaluated their association with clinical parameters and OS. This research demonstrates how radiomics can complement genomic analysis in precision oncology by providing an accessible imaging-based alternative for tumor profiling.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003ePatient cohort and data collection\u003c/h2\u003e\u003cp\u003eThe study cohort consisted of locally advanced (LA) HNSCC patients. Transcriptome sequencing data for a total of 506 HNSCC cases, including LA-HNSCC patients, along with clinical and follow-up data, were downloaded from The Cancer Genome Atlas (TCGA) database (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://portal.gdc.cancer.gov/\u003c/span\u003e\u003cspan address=\"https://portal.gdc.cancer.gov/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). Samples missing complete clinical information, those with survival times less than 30 days, or those that were not solid tumors or lacked sequencing data were excluded from the study. For tumor staging, the patients were classified based on the AJCC staging system, focusing specifically on those categorized as locally advanced (LA-HNSCC).\u003c/p\u003e\u003cp\u003eThe TCGA whole-exome somatic mutation data were obtained in Mutation Annotation Format (MAF) and processed using the VarScan software to calculate tumor mutational burden (TMB). Gene expression profiles in fragments per kilobase of transcript per million mapped reads (FPKM) format were converted to transcripts per million (TPM) format for subsequent analyses. Clinical data included age, sex, race, smoking history, drinking history, primary tumor site, human papillomavirus (HPV) infection status, histological differentiation, American Joint Committee on Cancer (AJCC) stage, survival time, and survival status. HPV infection positivity was defined as p16 protein positivity by immunohistochemistry or HPV positivity by fluorescence in situ hybridization.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eGene set variation analysis\u003c/h3\u003e\n\u003cp\u003eDifferentially expressed genes (DEGs) were identified using the R package \u003cem\u003elimma\u003c/em\u003e, with thresholds set at |log2 fold change| \u0026gt;1.0 and a false discovery rate (FDR)\u0026thinsp;\u0026lt;\u0026thinsp;0.05. A heatmap was generated using the \u003cem\u003epheatmap\u003c/em\u003e package to visualize gene expression differences. Functional enrichment analyses, including Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses, were conducted using the \u003cem\u003eclusterProfiler\u003c/em\u003e package, with significance set at p- and q-values\u0026thinsp;\u0026lt;\u0026thinsp;0.05.\u003c/p\u003e\n\u003ch3\u003eCalculation of TMB value\u003c/h3\u003e\n\u003cp\u003eTMB was defined as the total number of somatic coding mutations, including base substitutions, insertions, and deletions, detected per megabase of tumor DNA. Based on a previous study [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e], the TMB value for each sample was calculated as the total number of mutations divided by the length of the human exonic region (38 Mb). Synonymous mutations were excluded from the TMB calculation. Mutation data for HNSCC samples, downloaded from TCGA, were analyzed using the R package \u003cem\u003emaftools\u003c/em\u003e. The optimal TMB cut-off value was determined based on the area under the receiver operating characteristic (ROC) curve (AUC) for survival prediction using X-tile software. Patients were stratified into low- and high-TMB groups according to the median TMB value (4.2 mutations/Mb). TMB values were then merged with corresponding survival information by matching patient ID numbers. Kaplan\u0026ndash;Meier survival analysis was performed using R packages to evaluate the association between TMB and prognosis in HNSCC. Additionally, the relationship between TMB values and clinical characteristics was assessed.\u003c/p\u003e\n\u003ch3\u003eImage acquisition and segmentation\u003c/h3\u003e\n\u003cp\u003eA total of 159 LA-HNSCC cases with enhanced CT images in the arterial phase were obtained from The Cancer Imaging Archive (TCIA) database (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.cancerimagingarchive.net/\u003c/span\u003e\u003cspan address=\"https://www.cancerimagingarchive.net/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), specifically from the Head and Neck Squamous Cell Carcinoma (HNSCC) Imaging Dataset. Only arterial phase CT images were included in the analysis to ensure consistency in the vascular contrast enhancement, which is typically most relevant for assessing tumor characteristics in HNSCC. The decision to use only the arterial phase images was based on prior studies indicating that this phase provides the most accurate representation of tumor vascularization and tumor-tissue differentiation, which are crucial for radiomics feature extraction and TMB prediction. Other CT phases, such as venous or delayed phases, were not considered due to their lesser ability to delineate tumor boundaries and vascular features in the context of HNSCC. The images were processed and analyzed to create a radiomics model. Only patients who had pre-treatment CT scans were included in the study, and all included cases were confirmed to be locally advanced by clinical assessment and AJCC staging We excluded samples that were postoperative, had inferior image quality, or were devoid of sequencing data. The following conditions were defined as poor image quality: the presence of artifacts in the image (e.g., motion artifacts, metal artifacts caused by dentures) or missing/incomplete scanning of the tumor site. All data and images were anonymized and are publicly available, exempting them from ethical approval and informed consent. The TCIA dataset used in this study corresponds to the 'HNSCC collection,' which contains radiologic imaging data for patients with head and neck squamous cell carcinoma. Ethical approval was obtained from the relevant Institutional Review Board (IRB) or Ethics Committee. Since the data is publicly available and anonymized, informed consent was waived. All data used in this study complies with ethical guidelines for research.\u003c/p\u003e\u003cp\u003eIn this study, we concentrated on resampling to a consistent voxel size (1\u0026times;1\u0026times;1 mm\u0026sup3;) across all images to minimize variability arising from differences in scanning equipment, protocols, and lesion sizes. While resampling helps standardize the images, it is important to note that other preprocessing methods, such as image discretization or intensity normalization, can further enhance the robustness and reproducibility of extracted features. Image discretization, which involves transforming continuous pixel values into discrete levels, could potentially improve the stability of radiomic features by reducing the impact of noise and inconsistencies in image intensity. However, prior research in radiomics has shown that resampling alone can often provide sufficient improvement in feature reproducibility when combined with careful segmentation. Given the high quality of the CT images from the TCIA dataset and the specific focus on arterial-phase images, we chose to focus primarily on resampling as our preprocessing step. Further exploration of additional preprocessing techniques, such as image discretization, may be considered in future studies to evaluate their impact on the robustness of radiomic features.\u003c/p\u003e\u003cp\u003eAll CT images were acquired using multi-detector computed tomography (MDCT) scanners from different institutions contributing to the TCIA database. The scanning protocols varied slightly depending on the institution; however, all images were contrast-enhanced arterial phase scans with slice thicknesses ranging from 1 to 3 mm. The tube voltage was typically set between 100\u0026ndash;140 kVp, and the tube current ranged from 100\u0026ndash;400 mA, adjusted automatically by the scanner\u0026rsquo;s dose modulation system. The field of view (FOV) and matrix size were standardized as much as possible to maintain consistency in image quality. Images were reconstructed using standard convolution kernels to optimize contrast resolution while minimizing artifacts. These standardized imaging parameters ensured the reproducibility and comparability of the radiomics analysis.\u003c/p\u003e\n\u003ch3\u003eRadiomic Features Extraction\u003c/h3\u003e\n\u003cp\u003eAll images were resampled to a uniform voxel size of 1 \u0026times; 1 \u0026times; 1 mm\u0026sup3; to minimize variability caused by differences in scanning equipment, imaging protocols, and lesion size. Lesion segmentation was performed using 3D Slicer software (version 4.10.2; \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.slicer.org/\u003c/span\u003e\u003cspan address=\"https://www.slicer.org/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). Two experienced radiologists jointly identified lesion boundaries and delineated the volumes of interest (VOIs). Both radiologists were blinded to the clinical and pathological information throughout the segmentation process. Feature extraction was performed using Python\u0026rsquo;s open-source pyradiomics 3.0.1 package. A total of 1316 radiomic features were extracted from each CT image. These features comprised:\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eFirst-order statistics features (n\u0026thinsp;=\u0026thinsp;18): Describing the intensity distribution within the region of interest (ROI), including metrics such as mean, median, entropy, skewness, and kurtosis.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eShape-based (3D) features (n\u0026thinsp;=\u0026thinsp;14): Characterizing tumor geometry, such as volume, surface area, sphericity, elongation, and flatness.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eTexture features, including:\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eGray-Level Co-occurrence Matrix (GLCM) features (n\u0026thinsp;=\u0026thinsp;24)\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eGray-Level Run Length Matrix (GLRLM) features (n\u0026thinsp;=\u0026thinsp;16)\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eGray-Level Size Zone Matrix (GLSZM) features (n\u0026thinsp;=\u0026thinsp;16)\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eGray-Level Dependence Matrix (GLDM) features (n\u0026thinsp;=\u0026thinsp;14)\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eNeighboring Gray Tone Difference Matrix (NGTDM) features (n\u0026thinsp;=\u0026thinsp;5)\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003eWavelet-transformed features (n\u0026thinsp;=\u0026thinsp;1200): Derived by applying all possible high-pass (H) and low-pass (L) filter combinations to the original images, generating eight decomposed images per lesion.\u003c/p\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003eConstruction and evaluation of radiomics models for TMB prediction\u003c/h2\u003e\u003cp\u003eAfter extraction, the TCIA dataset was randomly divided into training and validation datasets in a 7:3 ratio, resulting in 112 individuals in the training dataset and 47 in the validation dataset. The division was performed randomly and was not based on TMB status, prognosis, or any clinical characteristics. The training dataset was used for feature selection and model development, while the validation dataset was used to assess model performance. A signature selection strategy was applied, which included the following steps: univariate Cox regression analysis, least absolute shrinkage and selection operator (LASSO)-Cox regression with 5-fold cross-validation, and multivariate Cox regression analysis. The optimal Lambda Lasso_output_min consists of features for establishing a radiomics signature models. Afterwards, clinical models and radiomics signature models were constructed and visualized using nomogram. The performance of the ROC curve validation model was assessed to evaluate the discriminative ability of the radiomics signature in predicting TMB status. The same methodology was used for both TMB and OS prediction models; however, each model targets a distinct clinical outcome, TMB and OS, requiring independent analysis and validation for each.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eConstruction and evaluation of radiomics models for OS prediction\u003c/h3\u003e\n\u003cp\u003eOS was defined as the duration from treatment initiation until any event resulting in patient death. If a patient remains alive at the last follow-up, the OS time is censored. Kaplan-Meier survival curves were used to estimate the OS rates, and both univariate and multivariate Cox proportional hazards regression analyses were performed to examine the determinants of OS. The same random 7:3 split used for the TMB prediction model was applied, with 112 individuals in the training dataset and 47 in the validation dataset. This division was not based on prognosis status or survival outcomes. Feature selection was performed using the training dataset, and model performance was evaluated on the validation dataset. We employed a signature selection strategy that consisted of the following steps: univariate Cox regression analysis, least absolute shrinkage and selection operator (LASSO)\u0026ndash;Cox regression with 5-fold cross-validation, and multivariate Cox regression analysis. The optimal Lambda Lasso_output_min consists of features for establishing a radiomics signature models. Afterwards, clinical models and radiomics signature models were constructed and visualized using nomogram. The performance of the ROC curve validation model was assessed to evaluate the discriminative ability of the radiomics signature in predicting TMB status. Although identical methods were employed to select features and construct models for TMB and OS, the distinct nature of TMB (a biomarker of mutation burden) and OS necessitates separate presentation of results for each model.\u003c/p\u003e\n\u003ch3\u003eClinical model construction\u003c/h3\u003e\n\u003cp\u003eThe clinical data included age, sex, histological differentiation, T stage, N stage, M stage, survival time, and survival status. Univariate logistic regression analysis and multivariate analysis were performed on clinical data to construct a clinical model for predicting TMB. Additionally, univariate and multivariate Cox proportional hazards regression analyses were conducted to identify factors associated with OS.\u003c/p\u003e\u003cp\u003eFor the TMB clinical model, univariate logistic regression analysis was used to assess the association between clinical variables and TMB status (high vs. low). Variables with a p-value\u0026thinsp;\u0026lt;\u0026thinsp;0.1 in the univariate analysis were entered into the multivariate logistic regression model to identify independent predictors and construct the final model.\u003c/p\u003e\u003cp\u003eFor the OS clinical model, univariate Cox proportional hazards regression analysis was performed to evaluate the prognostic significance of clinical variables for OS. Variables with a p-value\u0026thinsp;\u0026lt;\u0026thinsp;0.1 in the univariate analysis were included in the multivariate Cox regression to identify independent prognostic factors and develop the final model.\u003c/p\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003eStatistical Analysis\u003c/h2\u003e\u003cp\u003eStatistical analysis and model development were performed using R 3.3.3 software (IBM, Armonk, NY, USA). The standardized variables were expressed as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation (SD) or as median (25th percentile, 75th percentile), and evaluated using univariate logistic regression analysis, univariate Cox regression analysis, and multivariate Cox regression analysis.\u003c/p\u003e\u003cp\u003eOS was defined as the duration from randomization to the occurrence of any event resulting in patient death. OS rates for patients in both groups were assessed through Kaplan-Meier analysis. Both univariate and multivariate Cox proportional hazards regression analyses were performed to identify the determinants of OS. P\u0026thinsp;\u0026lt;\u0026thinsp;0.05 was considered statistically significant.\u003c/p\u003e\u003cp\u003eThe DeLong test was used to compare the AUC values between the training and validation datasets. ROC curves were plotted using the \"pROC\" package. Calibration curves were generated using the \"rms\" package, and decision curve analysis (DCA) was performed using the \"dca\" function.\u0026rdquo;\u003c/p\u003e\u003cp\u003eFor the TMB prediction task, classification methods (logistic regression and LASSO-based feature selection) were employed, while OS prediction used Cox proportional hazards regression appropriate for time-to-event data.\u003c/p\u003e\u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec13\"\u003e\n \u003ch2\u003e1 Mutations in HNSCC patients\u003c/h2\u003e\n \u003cp\u003eThe TCGA-HNSCC dataset includes data from 506 patients with head and neck squamous cell carcinoma. DEGs were identified using |log₂FC| \u0026gt;1.0 and an FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.05 as the cut-off criteria. A total of 136 genes were upregulated and 293 genes were downregulated (Fig. 1). To further investigate the biological significance of these DEGs, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were performed. GO enrichment analysis revealed that the DEGs were mainly involved in immune system processes, signaling receptor activity, and plasma membrane development (Fig. 2A\u0026ndash;C). KEGG enrichment analysis indicated that the DEGs were primarily enriched in cytokine\u0026ndash;cytokine receptor interaction, focal adhesion, and cell cycle pathways (Fig. 2D).\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec14\"\u003e\n \u003ch2\u003e2 Analysis of TMB and clinical relevance in HNSCC\u003c/h2\u003e\n \u003cp\u003eThe cut-off value for TMB was determined based on the AUC values for survival time using X-tile software. Patients were subsequently divided into low- and high-TMB groups, with a threshold of 4.2 mutations/Mb (Table 1). Analysis of somatic mutations revealed that the top 10 most frequently mutated genes in HNSCC were TP53 (70.3%), TTN (40.2%), FAT1 (23.4%), CDKN2A (20.8%), MUC16 (19.2%), CSMD3 (18.5%), PIK3CA (17.7%), NOTCH1 (17.5%), SYNE1 (17.1%), and LRP1B (16.4%) (Fig. 3A). Kaplan-Meier survival analysis demonstrated that patients in the high-TMB group had significantly worse OS compared to those in the low-TMB group (HR\u0026thinsp;=\u0026thinsp;1.65, 95% CI: 1.20\u0026ndash;2.28, p\u0026thinsp;=\u0026thinsp;0.002; Fig. 3B).\u003c/p\u003e\n \u003cdiv\u003e\u003cbr\u003e\u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec15\"\u003e\n \u003ch2\u003e3 Construction and evaluation of radiomics models for Prediction TMB\u003c/h2\u003e\n \u003cp\u003eA total of 159 HNSCC patients from the TCIA database were included in this study. To ensure unbiased validation, the dataset was randomly divided into a training cohort (n\u0026thinsp;=\u0026thinsp;112) and a validation cohort (n\u0026thinsp;=\u0026thinsp;47) at a 7:3 ratio. The training and validation cohorts were comparable in terms of clinical variables, molecular subtypes, and TMB status.\u003c/p\u003e\n \u003cp\u003eUnivariate COX regression analysis identified a total of 1316 features with statistically significant differences between the Low-TMB and High-TMB groups in the training set. Key features selected for the TMB prediction model included texture-based features, such as Gray-Level Co-occurrence Matrix (GLCM) entropy, which reflects the heterogeneity in the tumor\u0026apos;s intensity patterns, and volume-based shape features, involving tumor volume and surface area, which are critical indicators of tumor growth and aggressiveness. These features were associated with higher TMB levels, as increased tumor heterogeneity often correlates with a greater mutation burden. Using LASSO regression, the 1316 features were narrowed down to 16 key features. Notably, wavelet-transformed texture features, such as GLCM contrast and GLDM dependence variance, were selected as significant predictors. These features highlight the complexity and variation in tumor texture, which may reflect underlying biological differences in TMB, such as higher genomic instability and mutation rates (Fig. 4A, B). In the radiomics model for TMB prediction, key features, such as GLCM entropy, surface area-to-volume ratio, and wavelet-transformed contrast emerged as the most important predictors. Tumors with high entropy and contrast values tend to show more disordered texture and heterogeneity, characteristics that have been linked to higher mutation burden. These findings support the hypothesis that complex tumor textures are associated with greater genomic instability. The radiomics model showed strong discrimination in the training set (AUC\u0026thinsp;=\u0026thinsp;0.881); however, its validation set performance (AUC\u0026thinsp;=\u0026thinsp;0.64) indicates moderate predictive ability. When combined with clinical features, the model\u0026rsquo;s performance improved (AUC\u0026thinsp;=\u0026thinsp;0.902 in training and 0.669 in validation), suggesting added value in integrating radiomic and clinical features (Fig. 4C, D).\u003c/p\u003e\n \u003cp\u003eUnivariate and multivariate logistic regression analyses revealed that TMB status was significantly associated only with the primary tumor site (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Among different tumor sites, laryngeal cancer exhibited the highest TMB values, while oropharyngeal cancer showed the lowest. The clinical model achieved AUC values of 0.596 and 0.575 in the training and validation cohorts, respectively. In contrast, the combined model demonstrated improved predictive performance for TMB status, with AUC values of 0.902 in the training cohort and 0.669 in the validation cohort (Fig. 4C, D).\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec16\"\u003e\n \u003ch2\u003e5 Construction and evaluation of radiomics models for Prediction OS\u003c/h2\u003e\n \u003cp\u003eIn addition, radiomics models were utilized to predict OS in HNSCC patients. A total of 159 HNSCC patients from the TCIA database were included in this study. Based on the grouping used for TMB prediction models, feature selection was performed in the training dataset, which consisted of 112 patients, while 47 patients were allocated to the validation dataset.\u003c/p\u003e\n \u003cp\u003eUnivariate COX regression analysis identified key features associated with survival outcomes, including first-order features, such as mean intensity and skewness, which capture the overall distribution of voxel intensities within the tumor. Higher mean intensity and more positive skewness were correlated with worse prognosis, reflecting the presence of areas with more aggressive tumor growth. Additionally, shape features, including tumor sphericity and elongation were also important, as tumors with irregular shapes tend to exhibit more invasive behavior and poorer outcomes. The LASSO logistic regression model, with five-fold cross-validation under the minimum criterion, was applied to reduce the 1,316 features to 10 selected features and to construct the radiomics signature (Fig. 5A-B). The radiomics model for OS prediction identified first-order features such as mean intensity and skewness, along with shape features like sphericity. Higher mean intensity and less spherical tumors were associated with worse survival, suggesting that more irregular tumors may be more aggressive and less responsive to treatment. These results underscore the importance of tumor texture and shape in predicting long-term patient outcomes. The model\u0026rsquo;s AUC values in the training and validation sets were 0.817, 0.672, and 0.601, and 0.729, 0.715, and 0.711, respectively (Fig. 5C, D). Univariate and multivariate logistic regression analysis revealed that T stage and gender were significantly associated with worse OS (OS) (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05). The clinical model\u0026rsquo;s AUC values for the 1-, 3-, and 5-year survival rates in the training and validation sets were 0.626, 0.677, and 0.651, and 0.667, 0.607, and 0.583, respectively (Fig. 5E, F).\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec17\"\u003e\n \u003ch2\u003e6 Using Combined model to Prognosticate OS\u003c/h2\u003e\n \u003cp\u003eVariables that were significant in both univariable and multivariable analyses (T stage, gender, and radiomics features) were used to develop the combined model nomogram (Fig. 6A). After evaluation, the combined model demonstrated favorable predictive ability for OS, as indicated by the ROC curve for 1-, 3-, and 5-year survival rates. The model\u0026rsquo;s AUC values in the training and validation sets were 0.784, 0.693, and 0.665, and 0.772, 0.734, and 0.680, respectively (Fig. 6B, C). The Kaplan\u0026ndash;Meier survival curves for the combined model are shown in Figs. 6D and E. Furthermore, a high-risk combined model was significantly associated with lower OS in both the training and validation sets (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05).\u003c/p\u003e\n\u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eHNSCC is the sixth most common malignancy worldwide [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Despite significant advances in screening, diagnosis, and treatment, particularly with the advent of immunotherapy, HNSCC remains associated with poor prognosis [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Increasingly, cancer genomic data, including somatic mutations, copy number variations, and TMB, are being integrated to improve prognosis prediction and treatment response. A comprehensive analysis of 45 clinical studies involving 103,078 cancer patients suggested that high TMB may be an unfavorable prognostic indicator for patients receiving non-immunotherapy treatments. However, in patients treated with immunotherapy, high TMB has been globally associated with improved survival and better treatment efficacy [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eIn our study, we examined the prognostic significance of TMB and its correlation with immune cell infiltration in HNSCC, using the TCGA dataset. Our findings indicate that high-TMB status correlates with poor OS, which aligns with a previous multicenter retrospective study in Germany [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. In recent years, ICIs have shown promising therapeutic potential in recurrent/metastatic (R/M) HNSCC. Pembrolizumab and nivolumab, monoclonal antibodies targeting the PD-1 receptor, have demonstrated durable therapeutic efficacy in HNSCC [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. Studies have also indicated that TMB, particularly defined as the total number of somatic mutations per coding region of a tumor genome, can be a predictor of response to ICIs in R/M HNSCC [\u003cspan additionalcitationids=\"CR21 CR22 CR23\" citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. Patients with high TMB in the KEYNOTE-012 and KEYNOTE-055 trials showed efficacy in response to pembrolizumab, defined as \u0026ge;\u0026thinsp;10 mut/Mb [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. TMB has emerged as a potential biomarker for predicting immunotherapy response in prospective clinical trials of HNSCC.\u003c/p\u003e\u003cp\u003eHowever, the testing methods for TMB, such as next-generation sequencing (NGS) and whole exome sequencing (WES), are costly, time-consuming, and require fresh samples. The need for a more accessible, cost-effective alternative is evident. In this regard, CT radiomics has surfaced as a promising approach to identify potential immunotherapy responders and improve patient stratification for treatment decisions [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. With the rapid advancement of artificial intelligence (AI) and machine learning, these methods now offer substantial potential for integrating multi-omics datasets to improve clinical decision-making in HNSCC treatment [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e].\u003c/p\u003e\u003cp\u003ePrevious radiomics studies in head and neck tumors have shown the ability of radiomics models to predict various tumor characteristics, such as TMB and immunotherapy responses. For instance, CT-based radiomics models have successfully predicted TMB in cancers like non-small cell lung cancer and lower-grade gliomas [\u003cspan additionalcitationids=\"CR36\" citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]. However, no studies to date have used radiomics to predict TMB in HNSCC patients. In our study, a CT-based radiomics model was developed to predict TMB, demonstrating good performance in both the training and validation sets (AUC\u0026thinsp;=\u0026thinsp;0.881 and 0.64, respectively). These findings suggest that radiomics could be a valuable tool for TMB expression detection, particularly when biopsy is not feasible. While our model showed promising results, it is important to note that this radiomics signature model may complement, rather than replace, traditional methods for TMB detection, such as tissue biopsy and sequencing.\u003c/p\u003e\u003cp\u003eThe predictive ability of our model could also complement the TNM staging system. This model provides supplementary information that could influence treatment decisions, particularly in determining which patients may benefit from treatment deintensification. However, it is essential to consider that our study does not directly address how this model could change clinical decision-making or guide treatment strategies. While our radiomics-based approach may assist in identifying patients who are likely to benefit from immunotherapy, further investigation is required to explore its role in treatment personalization and clinical practice.\u003c/p\u003e\u003cp\u003eThere are some limitations to this study. Firstly, the use of data from public databases may introduce variability and bias, especially in the validation process. External cohort validation is essential to confirm the applicability of this model across different institutions and imaging systems. Secondly, manual lesion delineation introduces human error and variability, which could affect the reproducibility of the results. Future research should focus on developing automated, robust segmentation techniques to reduce these biases. Thirdly, overfitting remains a concern, particularly given the discrepancy in AUC values between the training (0.881) and validation sets (0.64). While the model shows good predictive ability, the lower performance in the validation set suggests the need for further refinement. Additionally, the comparison of radiomics-based approaches with other TMB prediction methods, such as circulating tumor DNA (ctDNA) or machine learning models using clinical features, was not explored in this study. Although the differences in clinical characteristics and prognosis between high- and low-TMB groups were analyzed using TCGA data, a similar analysis could not be performed in the TCIA dataset due to the limited availability of clinical and follow-up data. Instead, the TCIA data were utilized primarily for the development and validation of radiomics-based models for TMB and OS prediction. Despite the moderate AUC values in validation, the combined model showed a significant association with OS, suggesting its potential utility in clinical decision-making with further refinement. Future research should address this gap by comparing the effectiveness of these methods in predicting TMB and their potential to guide treatment strategies. To improve the generalizability of our findings, future studies will incorporate independent external validation cohorts and assess the models' performance across diverse populations. Further optimization of feature selection methods and model training strategies, such as deep learning approaches or ensemble models, may enhance predictive accuracy.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eIn conclusion, TMB status serves as a significant and independent prognostic factor for HNSCC. The combined model, based on enhanced CT images, demonstrates potential in predicting TMB status and shows promise for prognostic assessment in HNSCC. Leveraging this model may aid clinicians in making more informed decisions and support individualized and precise diagnosis and treatment strategies for HNSCC patients. However, further validation, particularly with external datasets, is necessary to confirm its generalizability and clinical utility.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePatient consent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data that support the fndings of this study are available from TCIA (https://www.cancerimagingarchive. net) and TCGA (https://portal.gdc.cancer.gov).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by University of Electronic Science and Technology, Sichuan Cancer Hospital Oncology Medical Industry Innovation Fund (ZYGX2021 YGCX006) , Sichuan Science and Technology Department Key Research and Development Project 2023YFS0157).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026apos; contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eYKL and NH designed the study. LL and YJQ conducted data analyses and drafted the manuscript.\u0026nbsp;MF, YQ, JRX,\u0026nbsp;and\u0026nbsp;JLR\u0026nbsp;contributed to the\u0026nbsp;data analyses. All authors\u0026nbsp;participated in the manuscript\u0026apos;s development\u0026nbsp;and approved the final version.\u0026nbsp;This\u0026nbsp;research\u0026nbsp;was supported by the Fundamental Research Funds for the Central Universities\u0026nbsp;(ZYGX2021YGCX006).\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eBray, F., Ferlay, J., Soerjomataram, I., et al. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. \u003cem\u003eCA Cancer J Clin.\u003c/em\u003e 2018;68(6):394\u0026ndash;424. https://doi.org/10.3322/caac.21492.\u003c/li\u003e\n\u003cli\u003eNCCN Clinical Practice Guidelines in Oncology (NCCN Guidelines\u0026reg;). Head and Neck Cancers Version 2.2023. https://www.nccn.org/professionals/physician_gls/pdf/head-and-neck.pdf.\u003c/li\u003e\n\u003cli\u003eBurtness, B., Harrington, K.J., Greil, R., et al. Pembrolizumab alone or with chemotherapy versus cetuximab with chemotherapy for recurrent or metastatic squamous cell carcinoma of the head and neck (KEYNOTE-048): a randomised, open-label, phase 3 study. \u003cem\u003eLancet.\u003c/em\u003e 2019;394(10212):1915\u0026ndash;1928. https://doi.org/10.1016/S0140-6736(19)32591-7.\u003c/li\u003e\n\u003cli\u003eCohen, E.E., Ferris, R.L., Psyrri, A., et al. Primary results of the phase III JAVELIN head \u0026amp; neck 100 trial: avelumab plus chemoradiotherapy (CRT) followed by avelumab maintenance vs CRT in patients with locally advanced squamous cell carcinoma of the head and neck (LA SCCHN). \u003cem\u003eAnn Oncol.\u003c/em\u003e 2020;31(Suppl 4):S658 . https://doi.org/10.1016/j.annonc.2020.08.1016.\u003c/li\u003e\n\u003cli\u003eLee, N.Y., Ferris, R.L., Psyrri, A., et al. Avelumab plus standard-of-care chemoradiotherapy versus chemoradiotherapy alone in patients with locally advanced squamous cell carcinoma of the head and neck: a randomised, double-blind, placebo-controlled, multicentre, phase 3 trial. \u003cem\u003eLancet Oncol.\u003c/em\u003e 2021;22(4):450\u0026ndash;462. https://doi.org/10.1016/S1470-2045(21)00076-0.\u003c/li\u003e\n\u003cli\u003eMachiels, J., Tao, Y., Burtness, B., et al. Primary results of the phase III KEYNOTE-412 study: pembrolizumab (pembro) with chemoradiation therapy (CRT) vs placebo plus CRT for locally advanced (LA) head and neck squamous cell carcinoma (HNSCC). \u003cem\u003eAnn Oncol.\u003c/em\u003e 2022;33(suppl_7):S808 \u0026ndash;S869. https://doi.org/10.1016/j.annonc.2022.07.714.\u003c/li\u003e\n\u003cli\u003eYilmaz, E., Ismaila, N., Dabney, R., et al. Immunotherapy and Biomarker Testing in Recurrent and Metastatic Head and Neck Cancers: ASCO Guideline Q and A. \u003cem\u003eJCO Oncol Pract.\u003c/em\u003e 2023;19(4):194\u0026ndash;196. https://doi.org/10.1200/OP.22.00789.\u003c/li\u003e\n\u003cli\u003eLitchfield, K., Reading, J.L., Puttick, C., et al. Meta-analysis of tumor- and T cell-intrinsic mechanisms of sensitization to checkpoint inhibition. \u003cem\u003eCell.\u003c/em\u003e 2021;184(3):596\u0026ndash;614.e14. https://doi.org/10.1016/j.cell.2021.01.002.\u003c/li\u003e\n\u003cli\u003eBera, K., Braman, N., Gupta, A., et al. Predicting cancer outcomes with radiomics and artificial intelligence in radiology. \u003cem\u003eNat Rev Clin Oncol.\u003c/em\u003e 2022;19(2):132\u0026ndash;146. https://doi.org/10.1038/s41571-021-00560-7.\u003c/li\u003e\n\u003cli\u003eMayerhoefer, M.E., Materka, A., Langs, G., et al. Introduction to Radiomics. \u003cem\u003eJ Nucl Med.\u003c/em\u003e 2020;61(4):488\u0026ndash;495. https://doi.org/10.2967/jnumed.119.232231.\u003c/li\u003e\n\u003cli\u003eZhu, Y., Wu, M. Noninvasive radiomic analysis of enhanced CT predicts CTLA4 expression and prognosis in head and neck squamous cell carcinoma. \u003cem\u003eSci Rep.\u003c/em\u003e 2023;13(1):16782. https://doi.org/10.1038/s41598-023-44055-0.\u003c/li\u003e\n\u003cli\u003eYe, J., Luo, J., Xu, S., et al. One-slice CT image based kernelized radiomics model for the prediction of low/mid-grade and high-grade HNSCC. \u003cem\u003eComput Med Imaging Graph.\u003c/em\u003e 2020;80:101675. https://doi.org/10.1016/j.compmedimag.2020.101675.\u003c/li\u003e\n\u003cli\u003eBruixola, G., et al. Radiomics and radiogenomics in head and neck squamous cell carcinoma: Potential contribution to patient management and challenges. \u003cem\u003eCancer Treat Rev.\u003c/em\u003e 2021;99:102263. https://doi.org/10.1016/j.ctrv.2021.102263.\u003c/li\u003e\n\u003cli\u003eYu, G., Wang, L.G., Han, Y., He, Q.Y. clusterProfiler: an R package for comparing biological themes among gene clusters. \u003cem\u003eOMICS.\u003c/em\u003e 2012;16(5):284\u0026ndash;287. https://doi.org/10.1089/omi.2011.0118.\u003c/li\u003e\n\u003cli\u003eLv, J., Zhu, Y., Ji, A., Zhang, Q., Liao, G. Mining TCGA database for tumor mutation burden and their clinical significance in bladder cancer. \u003cem\u003eBiosci Rep.\u003c/em\u003e 2020;40(4):BSR20194337 . https://doi.org/10.1042/BSR20194337.\u003c/li\u003e\n\u003cli\u003eCao, D., Xu, H., Xu, X., Guo, T., Ge, W. High tumor mutation burden predicts better efficacy of immunotherapy: a pooled analysis of 103078 cancer patients. \u003cem\u003eOncoimmunology.\u003c/em\u003e 2019;8(9):e1629258 . https://doi.org/10.1080/2162402X.2019.1629258.\u003c/li\u003e\n\u003cli\u003eEder, T., Hess, A.K., Konschak, R., et al. Interference of tumour mutational burden with outcome of patients with head and neck cancer treated with definitive chemoradiation: a multicentre retrospective study of the German Cancer Consortium Radiation Oncology Group. \u003cem\u003eEur J Cancer.\u003c/em\u003e 2019;116:67\u0026ndash;76. https://doi.org/10.1016/j.ejca.2019.04.015.\u003c/li\u003e\n\u003cli\u003eBurtness, B., et al. Pembrolizumab alone or with chemotherapy versus cetuximab with chemotherapy for recurrent or metastatic squamous cell carcinoma of the head and neck (KEYNOTE-048): a randomised, open-label, phase 3 study. \u003cem\u003eLancet.\u003c/em\u003e 2019;394(10212):1915\u0026ndash;1928. https://doi.org/10.1016/S0140-6736(19)32591-7.\u003c/li\u003e\n\u003cli\u003eFerris, R.L., et al. Nivolumab vs investigator\u0026apos;s choice in recurrent or metastatic squamous cell carcinoma of the head and neck: 2-year long-term survival update of CheckMate 141 with analyses by tumor PD-L1 expression. \u003cem\u003eOral Oncol.\u003c/em\u003e 2018;81:45\u0026ndash;51. https://doi.org/10.1016/j.oraloncology.2018.04.008.\u003c/li\u003e\n\u003cli\u003eZhang, Y., Lin, A., Li, Y., et al. Age and mutations as predictors of the response to immunotherapy in head and neck squamous cell cancer. \u003cem\u003eFront Cell Dev Biol.\u003c/em\u003e 2020;8:608969. https://doi.org/10.3389/fcell.2020.608969.\u003c/li\u003e\n\u003cli\u003eWildsmith, S., Li, W., Wu, S., et al. Tumor Mutational Burden as a Predictor of Survival with Durvalumab and/or Tremelimumab Treatment in Recurrent or Metastatic Head and Neck Squamous Cell Carcinoma. \u003cem\u003eClin Cancer Res.\u003c/em\u003e 2023;29(11):2066\u0026ndash;2074. https://doi.org/10.1158/1078-0432.CCR-22-2765.\u003c/li\u003e\n\u003cli\u003eChen, S., Yang, Y., He, S., et al. Review of biomarkers for response to immunotherapy in HNSCC microenvironment. \u003cem\u003eFront Oncol.\u003c/em\u003e 2023;13:1037884. https://doi.org/10.3389/fonc.2023.1037884.\u003c/li\u003e\n\u003cli\u003eValero, C., Golkaram, M., Vos, J.L., et al. Clinical-genomic determinants of immune checkpoint blockade response in head and neck squamous cell carcinoma. \u003cem\u003eJ Clin Invest.\u003c/em\u003e 2023;133(19):e169339 . https://doi.org/10.1172/JCI169339.\u003c/li\u003e\n\u003cli\u003eHaddad, R.I., Seiwert, T.Y., Chow, L.Q.M., et al. Influence of tumor mutational burden, inflammatory gene expression profile, and PD-L1 expression on response to pembrolizumab in head and neck squamous cell carcinoma. \u003cem\u003eJ Immunother Cancer.\u003c/em\u003e 2022;10(2):e003940 . https://doi.org/10.1136/jitc-2021-003940.\u003c/li\u003e\n\u003cli\u003ePfister, D.G., Haddad, R.I., Worden, F.P., et al. Biomarkers predictive of response to pembrolizumab in head and neck cancer. \u003cem\u003eCancer Med.\u003c/em\u003e 2023;12(6):6603\u0026ndash;6614. https://doi.org/10.1002/cam4.5478.\u003c/li\u003e\n\u003cli\u003eRodrigo, J.P., S\u0026aacute;nchez-Canteli, M., Otero-Rosales, M., et al. Tumor mutational burden predictability in head and neck squamous cell carcinoma patients treated with immunotherapy: systematic review and meta-analysis. \u003cem\u003eJ Transl Med.\u003c/em\u003e 2024;22(1):135. https://doi.org/10.1186/s12967-024-04935-z.\u003c/li\u003e\n\u003cli\u003eVerdegaal, E.M., de Miranda, N.F., Visser, M., et al. Neoantigen landscape dynamics during human melanoma-T cell interactions. \u003cem\u003eNature.\u003c/em\u003e 2016;536(7614):91\u0026ndash;95. https://doi.org/10.1038/nature18945.\u003c/li\u003e\n\u003cli\u003eRodrigo, J.P., et al. Tumor mutational burden predictability in head and neck squamous cell carcinoma patients treated with immunotherapy: systematic review and meta-analysis. \u003cem\u003eJ Transl Med.\u003c/em\u003e 2024;22(1):135. https://doi.org/10.1186/s12967-024-04935-z.\u003c/li\u003e\n\u003cli\u003eVan den Bossche, V., Zaryouh, H., Vara-Messler, M., et al. Microenvironment-driven intratumoral heterogeneity in head and neck cancers: clinical challenges and opportunities for precision medicine. \u003cem\u003eDrug Resist Updat.\u003c/em\u003e 2022;60:100806. https://doi.org/10.1016/j.drup.2022.100806.\u003c/li\u003e\n\u003cli\u003eDong, C., Zheng, Y.M., Li, J., et al. A CT-based radiomics nomogram for differentiation of squamous cell carcinoma and non-Hodgkin\u0026apos;s lymphoma of the palatine tonsil. \u003cem\u003eEur Radiol.\u003c/em\u003e 2022;32(1):243\u0026ndash;253. https://doi.org/10.1007/s00330-021-08143-x.\u003c/li\u003e\n\u003cli\u003eZheng, Y.M., Xu, W.J., Hao, D.P., et al. A CT-based radiomics nomogram for differentiation of lympho-associated benign and malignant lesions of the parotid gland. \u003cem\u003eEur Radiol.\u003c/em\u003e 2021;31(5):2886\u0026ndash;2895. https://doi.org/10.1007/s00330-020-07388-2.\u003c/li\u003e\n\u003cli\u003eZheng, Y.M., Li, J., Liu, S., et al. MRI-Based radiomics nomogram for differentiation of benign and malignant lesions of the parotid gland. \u003cem\u003eEur Radiol.\u003c/em\u003e 2021;31(6):4042\u0026ndash;4052. https://doi.org/10.1007/s00330-021-07796-y.\u003c/li\u003e\n\u003cli\u003eZhao, L., Gong, J., Xi, Y., et al. MRI-based radiomics nomogram may predict the response to induction chemotherapy and survival in locally advanced nasopharyngeal carcinoma. \u003cem\u003eEur Radiol.\u003c/em\u003e 2020;30(1):537\u0026ndash;546. https://doi.org/10.1007/s00330-019-06371-w.\u003c/li\u003e\n\u003cli\u003eGillies, R.J., Kinahan, P.E., Hricak, H. Radiomics: images are more than pictures, they are data. \u003cem\u003eRadiology.\u003c/em\u003e 2016;278(2):563\u0026ndash;577. https://doi.org/10.1148/radiol.2015151169.\u003c/li\u003e\n\u003cli\u003eWang, J., Wang, J., Huang, X., et al. CT radiomics-based model for predicting TMB and immunotherapy response in non-small cell lung cancer. \u003cem\u003eBMC Med Imaging.\u003c/em\u003e 2024;24(1):45. https://doi.org/10.1186/s12880-024-01216-5.\u003c/li\u003e\n\u003cli\u003eWen, Q., Yang, Z., Dai, H., et al. Radiomics Study for Predicting the Expression of PD-L1 and Tumor Mutation Burden in Non-Small Cell Lung Cancer Based on CT Images and Clinicopathological Features. \u003cem\u003eFront Oncol.\u003c/em\u003e 2021;11:620246. https://doi.org/10.3389/fonc.2021.620246.\u003c/li\u003e\n\u003cli\u003eYang, J., Shi, W., Yang, Z., et al. Establishing a predictive model for tumor mutation burden status based on CT radiomics and clinical features of non-small cell lung cancer patients. \u003cem\u003eTransl Lung Cancer Res.\u003c/em\u003e 2023;12(4):808\u0026ndash;823. https://doi.org/10.21037/tlcr-23-123.\u003c/li\u003e\n\u003cli\u003eLHT, Chu, N.T., Tran, T.O., et al. A Radiomics-Based Machine Learning Model for Prediction of Tumor Mutational Burden in Lower-Grade Gliomas. \u003cem\u003eCancers (Basel).\u003c/em\u003e 2022;14(14):3492. https://doi.org/10.3390/cancers14143492.\u003c/li\u003e\n\u003cli\u003eMa, T., Zhang, Y., Zhao, M., et al. A machine learning-based radiomics model for prediction of tumor mutation burden in gastric cancer. \u003cem\u003eFront Genet.\u003c/em\u003e 2023;14:1283090. https://doi.org/10.3389/fgene.2023.1283090.\u003c/li\u003e\n\u003cli\u003eLee, H., Moon, S.H., Hong, J.Y., et al. A Machine Learning Approach Using FDG PET-Based Radiomics for Prediction of Tumor Mutational Burden and Prognosis in Stage IV Colorectal Cancer. \u003cem\u003eCancers (Basel).\u003c/em\u003e 2023;15(15):3841. https://doi.org/10.3390/cancers15153841.\u003c/li\u003e\n\u003cli\u003eZheng, Y., Yuan, M., Zhou, R., et al. A computed tomography-based radiomics signature for predicting expression of programmed death ligand 1 in head and neck squamous cell carcinoma. \u003cem\u003eEur Radiol.\u003c/em\u003e 2022;32(8):5362\u0026ndash;5370. https://doi.org/10.1007/s00330-022-08618-5.\u003c/li\u003e\n\u003cli\u003eYoon, J., Suh, Y.J., Han, K., et al. Utility of CT radiomics for prediction of PD-L1 expression in advanced lung adenocarcinomas. \u003cem\u003eThorac Cancer.\u003c/em\u003e 2020;11(4):993\u0026ndash;1004. https://doi.org/10.1111/1759-7714.13352.\u003c/li\u003e\n\u003cli\u003ePatel, H., Vock, D.M., Marai, G.E., et al. Oropharyngeal cancer patient stratification using random forest based learning over high-dimensional radiomic features. \u003cem\u003eSci Rep.\u003c/em\u003e 2021;11:1\u0026ndash;11. https://doi.org/10.1038/s41598-021-84444-x.\u003c/li\u003e\n\u003cli\u003eTang, F.H., Chu, C.Y.W., Cheung, E.Y.W. Radiomics AI prediction for head and neck squamous cell carcinoma (HNSCC) prognosis and recurrence with target volume approach. \u003cem\u003eBJR Open.\u003c/em\u003e 2021;3(1):20200073. https://doi.org/10.1259/bjro.20200073.\u003c/li\u003e\n\u003cli\u003eCorino, V.D.A., Bologna, M., Calareso, G., et al. A CT-Based Radiomic Signature Can Be Prognostic for 10-Months Overall Survival in Metastatic Tumors Treated with Nivolumab: An Exploratory Study. \u003cem\u003eDiagnostics (Basel).\u003c/em\u003e 2021;11(6):1006. https://doi.org/10.3390/diagnostics11061006.\u003c/li\u003e\n\u003cli\u003eHan, K., Joung, J.F., Han, M., et al. Locoregional Recurrence Prediction Using a Deep Neural Network of Radiological and Radiotherapy Images. \u003cem\u003eJ Pers Med.\u003c/em\u003e 2022;12(2):256. https://doi.org/10.3390/jpm12020256.\u003c/li\u003e\n\u003cli\u003eBernatz, S., B\u0026ouml;th, I., Ackermann, J., et al. Radiomics for therapy-specific head and neck squamous cell carcinoma survival prognostication (part I). \u003cem\u003eBMC Med Imaging.\u003c/em\u003e 2023;23(1):71. https://doi.org/10.1186/s12880-023-01026-1.\u003c/li\u003e\n\u003cli\u003eMes, S.W., van Velden, F.H.P., Peltenburg, B., et al. Outcome prediction of head and neck squamous cell carcinoma by MRI radiomic signatures. \u003cem\u003eEur Radiol.\u003c/em\u003e 2020;30(11):6311\u0026ndash;6321. https://doi.org/10.1007/s00330-020-06971-x.\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Table 1","content":"\u003cp\u003eTable 1 is available in the Supplementary Files section.\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"enhanced-CT, radiomics, tumor mutational burden, head and neck squamous cell carcinoma","lastPublishedDoi":"10.21203/rs.3.rs-6921523/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6921523/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003ePurpose\u003c/h2\u003e\u003cp\u003eTumor mutation burden (TMB) has emerged as a promising biomarker for predicting immunotherapy response. This study aimed to evaluate the potential of a CT-based radiomics model to predict TMB status and overall survival (OS) in patients with head and neck squamous cell carcinoma (HNSCC).\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e\u003cp\u003eSomatic mutation and transcriptome data of 506 HNSCC cases were retrieved from The Cancer Genome Atlas (TCGA) to calculate TMB and identify differentially expressed genes (DEGs). Functional enrichment analysis was conducted using Gene Ontology (GO) and KEGG databases. Kaplan-Meier and Cox regression analysis was used to assess the prognostic value of TMB. A cohort of 159 patients with pre-treatment contrast-enhanced CT scan from The Cancer Imaging Archive (TCIA) was used to extract radiomics features. The dataset was split into training (n\u0026thinsp;=\u0026thinsp;112) and validation (n\u0026thinsp;=\u0026thinsp;47) datasets. Feature selection was performed using univariate Cox regression and LASSO, followed by multivariate Cox analysis. Predictive models (radiomics, clinical, and combined) were evaluated using receiver operating characteristic (ROC) curves and area under the curve (AUC) values.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e\u003cp\u003eHigh TMB was associated with poor OS. Functional enrichment revealed that DEGs were enriched in immune processes and cell signaling. Notably, 16 features were selected for the TMB prediction model. The combined model achieved AUC values of 0.902 (training) and 0.669 (validation) for TMB prediction. For 1-, 3-, and 5-year OS prediction, the combined model achieved AUC values of 0.665\u0026ndash;0.784 in the training cohort and 0.680\u0026ndash;0.772 in the validation cohort.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e\u003cp\u003eThe CT-derived radiomics model demonstrated potential for noninvasive prediction of TMB and survival in HNSCC, appearing advantageous for immunotherapy decision-making.\u003c/p\u003e","manuscriptTitle":"CT Radiomics-Clinical Model for Noninvasive Prediction of Tumor Mutation Burden and Survival in Locally Advanced Head and Neck Squamous Cell Carcinoma","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-22 16:21:10","doi":"10.21203/rs.3.rs-6921523/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"1d08ae63-f4a4-4c51-91e2-d3e1774c1d7a","owner":[],"postedDate":"July 22nd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-08-12T08:38:45+00:00","versionOfRecord":[],"versionCreatedAt":"2025-07-22 16:21:10","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6921523","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6921523","identity":"rs-6921523","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}