Support Tools in the Differential Diagnosis of Salivary Gland Tumors through Inflammatory Biomarkers and Radiomics Metrics: A Preliminary Study

Simple Summary The management of salivary gland tumors (SGTs), especially their early diagnosis, remains a challenge for physicians. Indeed, differentiating benign and malignant SGTs is an essential step in choosing an appropriate surgical approach. The aim of this study was to increase the effectiveness of pre-surgical diagnosis through a machine learning (ML) diagnostic tool that evaluates inflammatory biomarkers and radiomic metrics extracted from magnetic resonance imaging (MRI) sequences. Specifically, we considered the following indices of inflammation as inflammatory biomarkers: the systemic immune-inflammation index (SII), the systemic inflammation response index (SIRI), the platelet-to-lymphocyte ratio (PLR), and the neutrophil-to-lymphocyte ratio (NLR). In the context of cancer research, however, radiomics enables high-performance quantitative analysis of radiological images. We concluded that inflammatory biomarkers and radiomic features are comparably capable of supporting a differential diagnosis and are easily obtained through the preclinical investigations of patients. Background: The purpose of this study was to investigate how the systemic inflammation response index (SIRI), systemic immune-inflammation index (SII), neutrophil/lymphocyte ratio (NLR) and platelet/lymphocyte ratio (PLR), and radiomic metrics (quantitative descriptors of image content) extracted from MRI sequences by machine learning increase the efficacy of proper presurgical differentiation between benign and malignant salivary gland tumors. Methods: A retrospective study of 117 patients with salivary gland tumors was conducted between January 2015 and November 2022. Univariate analyses with nonparametric tests and multivariate analyses with machine learning approaches were used. Results: Inflammatory biomarkers showed statistically significant differences (p < 0.05) in the Kruskal-Wallis test based on median values in discriminating Warthin tumors from pleomorphic adenoma and malignancies. The accuracy of NLR, PLR, SII, and SIRI was 0.88, 0.74, 0.76, and 0.83, respectively. Analysis of radiomic metrics to discriminate Warthin tumors from pleomorphic adenoma and malignancies showed statistically significant differences (p < 0.05) in nine radiomic features. The best multivariate analysis result was obtained from an SVM model with 86% accuracy, 68% sensitivity, and 91% specificity for six features. Conclusions: Inflammatory biomarkers and radiomic features can comparably support a pre-surgical differential diagnosis.