A nomogram for predicting the diagnosis of central malignant tumors based on preoperative clinical characteristics and laboratory indicators: a diagnostic study

Purpose: This study aimed to develop a nomogram to predict the preoperative diagnostic probabilities of central lymphoma and glioma, as well as glioblastoma versus non-glioblastoma. Patients and methods: Retrospective analysis was performed on patients with central nervous system lymphoma or glioma who received treatment at our department, between 2016 and 2025. From 2016 to 2024, Eligible patients were randomly assigned to training and validation sets in a 7:3 ratio. Patients at our department from 2024 to 2025 ( n = 104) and two other medical centers (External Center1: n = 95, External Center2: n = 123) will be included as prospective external validation cohorts. Key variables for nomogram construction were identified through the integration of least absolute shrinkage and selection operator (LASSO) and multivariate logistic regression analysis. To assess the performance of the nomogram, seven machine learning models were constructed, including logistic regression, decision tree, random forest, support vector machine (SVM), neural network, Extreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (lightGBM), which were then evaluated using the area under the receiver operating characteristic curve (AUC) and decision curve analysis. The validation set was used for further model evaluation. Results: This retrospective study included 712 cases: 127 patients with newly diagnosed primary central nervous system lymphoma (PCNSL) and 586 patients with newly diagnosed glioma. In the diagnostic model for PCNSL versus glioma, the following five risk factors were included: age, Karnofsky Performance Status (KPS), neutrophil count (NEUT), neutrophil ratio (NEUT1), and monocyte count (MONO). The area under the curve (AUC) for the seven models ranged from 0.784 to 0.889, and the optimal AUC values obtained from the external validation sets at our center (2024-2025) and two other medical centers were 0.877, 0.716, and 0.743, respectively. In the diagnostic model for glioblastoma versus non-glioblastoma, three risk factors were included: age, neutrophil ratio (NEUT1), and monocyte count (MONO). The AUC for the seven models ranged from 0.778 to 0.857, while the optimal AUC values obtained from the external validation sets at our center (2024-2025) and two other medical centers were 0.861, 0.842, and 0.710. Conclusion: This study developed and validated diagnostic probability models for central lymphoma versus glioma, and glioblastoma versus non-glioblastoma. These models may assist clinicians in determining the type of central malignant tumor affecting patients, thereby facilitating the development of more personalized and optimized treatment strategies.