Contrast-enhanced image-guided learning for nasopharyngeal carcinoma diagnosis using non-contrast MRI

Contrast-enhanced T1-weighted magnetic resonance imaging (MRI) using gadolinium-based contrast agents (GBCAs) offers exceptional differentiation between nasopharyngeal carcinoma (NPC) and non-malignant cases, making it indispensable for NPC diagnosis. However, safety concerns about GBCAs have led to restrictions on some linear-GBCAs in Europe and ongoing efforts to find safer alternatives. In this study, we developed a diagnostic model for NPC using non-contrast images, leveraging contrast-enhanced T1-weighted images for guidance during training. The model was trained on a dataset of 694 cases with an equal distribution of NPC and non-NPC cases (50%). Testing was performed on an independent dataset of 263 cases, consisting of 43% NPC and 57% non-NPC cases. The proposed model achieved high diagnostic accuracy, with performance comparable to models incorporating both contrast-enhanced and non-contrast images. These results highlight its potential as a viable contrast-free alternative for NPC diagnosis.