Background: Stage T1 nasopharyngeal carcinoma (NPC T1 ) and benign hyperplasia (BH) are 2 common causes of nasopharyngeal mucosa/submucosa thickening without specific clinical symptoms. The treatment management of these 2 entities is significantly different. Reliable differentiation between the 2 entities is critical for the treatment decision and prognosis of patients. Therefore, our study aims to explore the optimal energy level of noise-optimized virtual monoenergetic images [VMI (+)] derived from dual-energy computed tomography (DECT) to display NPC T1 and BH and to explore the clinical value of DECT for differentiating these 2 diseases.Methods: A total of 91 patients (44 NPC T1 , 47 BH) were enrolled. The demarcation of the lesion margins and overall image quality, noise, contrast-to-noise ratio (CNR), and signal-to-noise ratio (SNR) were evaluated for 40-80 kiloelectron volts (keV) VMIs (+) and polyenergetic images in the contrast-enhanced phase. Image features were assessed in the contrast-enhanced images with optimal visualization of NPC T1 and BH. The demarcation of NPC T1 and BH in iodine-water maps was also assessed. The contrast-enhanced images were used to calculate the slope of the spectral Hounsfield unit curve (λ HU ) and normalized iodine concentration (NIC). The nonenhanced phase images were used to calculate the normalized effective atomic number (NZ eff ).The attenuation values on 40-80 keV VMIs (+) in the contrast-enhanced phase were recorded. The diagnostic performance was assessed using receiver operating characteristic (ROC) curve analysis.Results: The 40 keV VMI (+) in the enhanced phase yielded higher demarcation of the lesion margins scores, overall image quality scores, noise, SNR, and CNR values than 50-80 keV VMIs (+) and polyenergetic images. NPC T1 yielded higher attenuation values on VMI (+) at 40 keV (A 40 ), NIC, λ HU , and NZ eff values than BH. The multivariate logistic regression model combining image features (tumor symmetry) with quantitative parameters (A 40 , NIC, λ HU , and NZ eff ) yielded the best performance for differentiating the 2 diseases (AUC: 0.963, sensitivity: 89.4%, specificity: 93.2%).
Conclusions:The combination of DECT-derived image features and quantitative parameters contributed to the differentiation between NPC T1 and BH.