Abstract
BACKGROUND: Annually, more than 550,000 people are diagnosed with head and neck squamous cell carcinoma (HNSCC) using invasive techniques, emphasizing the need for non-invasive diagnostic methods. Thus, our investigation aimed to create radiomics models that could forecast patients' MMP13 expression levels. METHODS: This study was based on downloading genomic data and enhanced computed tomography (CT) images of HNSCC patients from The Cancer Imaging Archive (TCIA) and The Cancer Genome Atlas (TCGA) databases for prognostic analyses, image feature extraction, and construction of the radiomics models. Survival analysis (Kaplan-Meier survival curve, COX regression analysis, subgroup analysis, and interaction test) was used to explore the prognostic value of MMP13 in HNSCC. Correlation analysis, differential analysis of immune cell infiltration and enrichment analysis were used to explore the potential molecular mechanisms of CXCL8 expression and its relationship with the immune microenvironment. Radiomics models were constructed and Rad-score-based prediction of MMP13 and epithelial-mesenchymal transition (EMT) expression in HNSCC tissues was performed. Finally, intraclass correlation coefficient (ICC) was used to evaluate the consistency of radiomics features. RESULTS: TCGA had a total of 483 HNSCC patients, of which high (n=326) and low (n=157) MMP13-expressing groups were used in the survival analysis. The MMP13 high-expression and low-expression groups had respective median survival times of 36.43 and 65.73 months. Tumors expressed MMP13 at a substantially greater level than normal tissue. We established the Gradient Boosting Machine (GBM) model and the logistic regression (LR) model, respectively. The area under the curve (AUC) value in the training set was 0.864 (GBM) and 0.746 (LR), and in the validation set was 0.79 (GBM) and 0.73 (LR). The Hosmer-Lemeshow goodness-of-fit test and calibration curve both showed consistency (P>0.05) between the true and predicted values in GBM model. The decision curve analysis (DCA) display model exhibited good clinical practicality in both models. CONCLUSIONS: A noteworthy association was observed between MMP13 expression and the prediction of HNSCC. Predicting MMP13 expression levels may be accomplished with the use of radiomics, which is based on contrast-enhanced computed tomography (CECT).