Accurate Classification of Parotid Tumors Based on Apparent Diffusion Coefficient

Abstract

Purpose: In this work, we aimed to propose an automatic classification scheme based on the parameters derived from apparent diffusion coefficient (ADC)-maps for discriminating benign and malignant parotid tumors.
Methods: MRI was carried out prospectively on 41 patients presented with parotid tumors who underwent surgery and post-surgical histopathological assessment was provided for them (32 benign, 9 malignant). Based on anatomical images, regions of interest (ROIs) were selected on the most solid parts of tumors on ADC-maps. Three quantitative parameters, namely ADC-Mean, ADC-Max and ADC-Mean were calculated. Automatic classification of parotid tumors using ADC parameters was performed and assessed employing two different classifiers, namely, linear discriminant analysis (LDA) and quadratic discriminant analysis (QDA).
Results: Following statistical analysis, it was indicated that the ADC values in benign tumors are significantly higher than malignant tumors. ADC-Mean, and -Max presented statistically significant differences among benign and malignant parotid tumors (p<0.05). Among the extracted parameters, ADC-Max is the most relevant quantitative parameter for tumor classification with 82.9% accuracy, 84.4% specificity, 77.8% sensitivity, and 83.3% area under the ROC curve (AUC) by exploiting each of the automatic classifiers. This implies that this parameter is inherently accurate and adding further classification complexity does not improve the results. A linear classifier using LDA classification based on ADC-max is proposed, which indicates that ADC-Max under 1.48×10^-3mm²/s is highly suggestive of malignancy (with 83% accuracy).  
Conclusion: ADC-Max is a potential biomarker for discriminating benign and malignant parotid tumors. Using ADC-Max and LDA, a simple and clinically-feasible classifier is proposed.