Magnetic resonance image biomarkers improve differentiation of benign and malignant parotid tumors through diagnostic model analysis

Abstract

Objectives: To explore the effectiveness of magnetic resonance image (MRI)-based biomarkers for identifying benign and malignant parotid tumors via diagnostic model analysis.

Methods: This retrospective study included 109 patients (development cohort and validation cohort) who underwent MRI preoperatively, including T1- and T2-weighted images. Parameters based on 2D or 3D texture analysis were extracted from tumor lesions by MaZda software, fisher discriminant and bootstrap method were used to perform parameter reduction, diagnostic models with the selected biomarkers were established along with clinical data, model performance (discrimination and calibration) was furtherly evaluated by internal and external validation, decision curve analysis was applied to measure the improvement of clinical benefits.

Results: S(5,5) Entrop, S(0,1) ASM, WavEnHH (s-4), S(1,1,0) Entropy and Perc.10% were significantly associated with the pathological diagnosis of parotid tumor (benign versus malignancy), when adding these biomarkers to the regression analysis, model performance significantly improved in the development cohort (likelihood-ratio-test; p < 0.05, with an increase of AUC from 0.72 (reference model) to 0.85), and these results were maintained in a small external validation cohort. Decision curve analysis indicated that clinical benefit was greater with the application of MRI-based biomarkers.

Conclusions: MRI-based texture analysis is proven to be an effective tool in differentiating benign and malignant parotid tumors, preoperative diagnosis was improved with the selected biomarkers compared to the reference model.

Keywords: Differentiation; Magnetic resonance imaging; Parotid tumor; Radiomics; Texture analysis.