Prediction of Prognosis of Tongue Squamous Cell Carcinoma Based on Clinical MR Imaging Data Modeling

Abstract

Objective: Tongue squamous cell carcinoma (TSCC) is one of the most common and poor prognosis head and neck tumors. The purpose of this study is to establish a model for predicting TSCC prognosis based on clinical and MR radiomics data and to develop a nomogram. Methods: A retrospective analysis was performed on the clinical and imaging data of 211 patients with pathologically confirmed TSCC who underwent radical surgery at xx hospital from February 2011 to January 2020. Patients were divided into a study group (recurrence, metastasis, and death, n = 76) and a control group (normal survival, n = 135) according to 1 to 6 years of follow-up. A training set and a test set were established based on a ratio of 7:3 and a time point. In the training set, 3 prediction models (clinical data model, imaging model, and combined model) were established based on the MR radiomics score (Radscore) combined with clinical features. The predictive performance of these models was compared using the Delong curve, and the clinical net benefit of the model was tested using the decision curve. Then, the external validation of the model was performed in the test set, and a nomogram for predicting TSCC prognosis was developed. Results: Univariate analysis confirmed that betel nut consumption, spicy hot pot or pickled food, unclean oral sex, drug use, platelet/lymphocyte ratio (PLR), neutrophil/lymphocyte ratio (NLR), depth of invasion (DOI), low differentiation, clinical stage, and Radscore were factors that affected TSCC prognosis (P < .05). In the test set, the combined model based on these factors had the highest predictive performance for TSCC prognosis (area under curve (AUC) AUC: 0.870, 95% CI [0.761-0.942]), which was significantly higher than the clinical model (AUC: 0.730, 95% CI [0.602-0.835], P = .033) and imaging model (AUC: 0.765, 95% CI [0.640-0.863], P = .074). The decision curve also confirmed the higher clinical net benefit of the combined model, and these results were validated in the test set. The nomogram developed based on the combined model received good evaluation in clinical application. Conclusion: MR-LASSO extracted texture parameters can help improve the performance of TSCC prognosis models. The combined model and nomogram provide support for postoperative clinical treatment management of TSCC.

Keywords: combined model; least absolute shrinkage and selection operator; nomogram; prediction model; prognosis; radiomics; tongue squamous cell carcinoma.

Figure 1.

The results of quantitative literature analysis based on tongue squamous cell carcinoma (TSCC) search indicate that surgical treatment and molecular pathology has always been a hot topic in TSCC research, but clinical imaging radiomics prediction models have not been reported before.

Figure 2.

The inclusion and exclusion criteria and case grouping method of this study.

Figure 3.

The simple schematic diagram of tongue squamous cell carcinoma (TSCC) MR lesion delineation and radiomics extraction in this study.

Figure 4.

Male, 59 years old, with poorly differentiated tongue squamous cell carcinoma (TSCC). Pathological findings suggested that the tumor tissue was squamous differentiated and lacks of keratinizing beads (a). Female, 62 years old, with highly differentiated tongue squamous cell carcinoma. Pathology suggested squamous differentiation of the tumor tissue, with obvious keratinized beads visible (b).

Figure 5.

Delong nonparametric curves of the training set (a) and the test set (b). The area under the ROC curve of the combined model of the 2 groups is the largest, which confirms that the combined model has the best predictive performance. Clinical data model (A); imaging model (B); combined model (C).

Figure 6.

The higher clinical net benefits of the combined model was confirmed in the 2 groups by decision curve analysis (DCA) of training set (a) and test set (b) using R software.

Figure 7.

The nomogram prediction tool based on the risk factors of the combined model was used clinically (a, nomogram; b, Calibration). Namely, each risk factor is scored, added together, and the final risk value is calculated.