Combination of ultrasonography and MRI for preoperative prediction of lymph node metastasis in tongue squamous cell carcinoma: An exploratory study

Abstract

Preoperative depth of invasion (DOI) is a critical predictor of cervical lymph node metastasis (CLNM) in tongue squamous cell carcinoma (TSCC). Ultrasonography (US) offers high accuracy for shallow tumors, whereas magnetic resonance imaging (MRI) provides superior visualization of deeper structures. However, each modality has limitations. This retrospective, exploratory study aimed to investigate a combined strategy leveraging the complementary strengths of both modalities to improve preoperative CLNM prediction. The study included 46 patients with TSCC who underwent radical surgery between September 2014 and August 2019. Correlations between US-derived DOI (usDOI), MRI-derived DOI (mrDOI), and pathological DOI (pDOI) were assessed using Pearson’s product-moment correlation. Agreement was evaluated by Bland–Altman analysis. A grid search approach identified the optimal usDOI threshold for switching to mrDOI, and predictive performance for CLNM was evaluated using receiver operating characteristic (ROC) analysis with the area under the curve (AUC). Both usDOI and mrDOI showed strong correlations with pDOI (r = 0.956 and r = 0.958, respectively). However, usDOI demonstrated smaller measurement bias (mean difference: 0.88 mm) compared to mrDOI (2.43 mm). Bland–Altman plots revealed that usDOI tended to underestimate pDOI when the mean DOI exceeded 10 mm. Grid search analysis determined an optimal threshold of 7 mm for switching from US to MRI measurements. The combined DOI strategy —using usDOI for DOI ≤ 7 mm and mrDOI for DOI > 7 mm— yielded a numerically higher AUC (0.694) compared to either modality alone (usDOI: AUC = 0.682; mrDOI: AUC = 0.673). This study demonstrates the potential value of a conditional imaging strategy that applies US for shallow lesions and MRI for deeper ones. By capitalizing on the strengths of each modality, this approach offers a logical framework for improving the accuracy of preoperative CLNM prediction in TSCC. Larger, prospective studies are warranted to validate these findings.