18F-FDG PET/CT-Based Radiomics Nomogram for Prediction of Bone Marrow Involvement in Pediatric Neuroblastoma: A Two-Center Study

Abstract

To assess the predictive ability of an 18F-FDG PET/CT-based radiomics nomogram for bone marrow involvement in pediatric neuroblastoma. A total of 241 neuroblastoma patients who underwent 18F-FDG PET/CT at two medical centers were retrospectively evaluated. Data from center A (n=200) were randomized into a training cohort (n=140) and an internal validation cohort (n=60), while data from center B (n=41) constituted the external validation cohort. For each patient, two regions of interest were defined using the tumor and axial skeleton. The clinical factors and radiomics features were derived to construct the clinical and radiomics models. The radiomics nomogram was built by combining clinical factors and radiomics features. The area under the receiver operatingcharacteristic curves (AUCs) were used to assess the performance of the models. Radiomics models created from tumor and axial skeleton achieved AUCs of 0.773 and 0.900, and the clinical model had an AUC of 0.858 in the training cohort. By incorporating clinical risk factors and axial skeleton-based radiomics features, the AUC of the radiomics nomogram in the training cohort, internal validation cohort, and external validation cohort was 0.932, 0.887, and 0.733, respectively. The axial skeleton-based radiomics model performed better than the tumor-based radiomics model in predicting bone marrow involvement. Moreover, the radiomics nomogram showed that combining axial skeleton-based radiomics features with clinical risk factors improved their performance.