Related Experiment Video
Updated: Aug 25, 2026
Comparison of Predictive Performance of Three Lymph Node Staging Systems in Colorectal Signet Ring Cell Carcinoma Based on Machine Learning Model
Published on: April 18, 2025
Development and Validation of an MR Signal Intensity Ratio-Based Discriminant Classification Model for
Wen Li1,2, Ying-Qi Li2, Ge-Hua Tong3
1School of Chinese Medicine, Guangdong Pharmaceutical University, Guangzhou, China.
Background:
Early-stage Nasopharyngeal Carcinoma (NPC) and Nasopharyngeal Lymphoid Hyperplasia (NPLH) share similar clinical and MRI features, complicating non-invasive diagnosis. As MRI accuracy decreases with diffuse NPC signals, objective differentiation methods are urgently needed. Therefore, this study aims to evaluate the efficiency of an MRI Signal Intensity Ratio (SIR) based discriminant classification model in distinguishing NPLH from NPC.
Materials And Methods:
One hundred twenty-six patients with pathologically confirmed nasopharyngeal lesions (27 with NPLH and 99 with NPC) were retrospectively enrolled. The maximum, minimum, and mean signal intensities of the lesions were measured on T2-weighted imaging (T2WI) and post-contrast T1-weighted imaging (T1WI+C). Signal Intensity Ratios (SIRs) were calculated by normalizing lesion SI to that of the longus capitis muscle. A discriminant model was then developed to differentiate NPLH from NPC.
Results:
Stepwise discriminant analysis identified T2WI-Max-SIR and T1WI+C-Min-SIR as the optimal predictors. The established discriminant classification model yielded an original accuracy of 95.2% and a cross-validated accuracy of 93.7% in distinguishing NPLH from NPC.
Discussion:
NPC exhibited a significantly lower T2WI-SIR than NPLH. Incorporating these quantitative metrics into a discriminant model provides an objective assessment, minimizing scanner inconsistencies and overcoming the limitations of subjective visual evaluation.
Conclusion:
Integrating T2WI-Max-SIR and T1WI+C-Min-SIR into a discriminant classification model enables accurate, objective, and non-invasive differentiation between NPLH and NPC, which potentially reduces unnecessary biopsies.