Mô hình Nomogram cộng hưởng từ đa thông số trong phân biệt u thần kinh đệm bậc cao và u lympho não nguyên phát

Abstract

This study aimed to evaluate the potential of Diffusion Tensor Imaging (DTI), Magnetic Resonance Spectroscopy (MRS), and Magnetic Resonance Perfusion (MRP) parameters, and to develop a Nomogram model for differentiating High-Grade Gliomas (HGG) from Primary Central Nervous System Lymphomas (PCNSL). A retrospective study was conducted on 24 HGG and 23 PCNSL cases. Quantitative measurements of FA, MD, rCBV, rCBF, and Cho/NAA, Cho/Cr ratios were performed in the solid tumor (t) and peritumoral (p) regions. Optimal variables were screened using the LASSO algorithm. The Nomogram was constructed via Logistic Regression and internally validated using the Bootstrapping technique. The results identified three core variables selected for the Nomogram as MDt, rCBVt, and rCBVp. The model achieved an Area Under the Curve (AUC) of 0.978, showing high consistency on the Calibration curve with an extremely low Bootstrap-corrected Mean Absolute Error (MAE) of 0.049. The combination of MDt, rCBVt, and rCBVp through the LASSO-Nomogram machine learning model initially establishes a visual predictive tool to support the differential diagnosis of HGG and PCNSL.