Identification of primary central nervous system lymphoma from high-grade glioma based on a <SUP align="right">18</SUP>F-FDG PET/CT radiomics nomogram compared with deep learning: a multicentre study

Diagnosing central nervous system space-occupying lesions still depend on stereotactic biopsy. Therefore, a non-invasive imaging method is urgently required to distinguish between the two. This retrospective study enrolled 66 patients (38 with PCNSL and 28 with HGG) who underwent 18F-fluorodeoxyglucose-positron emission tomography/computed tomography (18F-FDG PET/CT) between July 2017 and July 2022 to investigate the ability of 18F-FDG PET/CT -based radiomics features in differentially diagnosing PCNSL and HGG. A group of 40 patients was assigned as the training cohort, while another group of 26 patients as the validation cohort. A total of 788 radiomics features were extracted from 18F-FDG PET/CT images in the training cohort. Two features were selected by the LASSO method from 788 features to build the logistic model and radiomics nomogram. The AUC of the radiomics nomogram for discriminating PCNSL from HGG was 0.960 [95% confidence interval (CI): 0.909-1] and 0.920 (95% CI: 0.794-1) in the training and validation cohorts, respectively. The training and validation revealed that the established radiomics nomogram model of 18F-FDG PET/CT displayed excellent discrimination capabilities in PCNSL and HGG, and may have the potential to improve diagnostic accuracy and patient outcomes.