Fusion of Intratumoral and Peritumoral Ultrasound Radiomics with Clinical Features for Distinguishing Diffuse Large B-cell Lymphoma from Hodgkin Lymphoma

Objectives To develop a model integrating clinical features with intratumoral and peritumoral ultrasound radiomics features for distinguishing Diffuse Large B-Cell Lymphoma (DLBCL) from Hodgkin Lymphoma (HL). Methods This study retrospectively analyzed data from 102 DLBCL patients (218 enlarged lymph nodes) and 37 HL patients (105 enlarged lymph nodes) at The First Affiliated Hospital of Chongqing Medical University. The enlarged lymph nodes were randomly allocated to a training cohort (227 nodes) and a test cohort (96 nodes) in a 7:3 ratio. Radiomics models were constructed based on ultrasound images from both the primary tumor and the peritumoral area. For the optimally performing regions-of-interest, two fusion strategies, a feature-based and a decision-based model, were applied to build the fusion models. The performance of each model was assessed using the area under the receiver operating characteristic curve (AUC) and decision curve analysis (DCA). The study strictly adhered to the TRIPODAI checklist. Results In the test cohort, the radiomics model using a 16-pixel peritumoral extension demonstrated optimal performance with an AUC of 0.636. The decision-based fusion model (Radexpand16_Clinica_R) achieved the highest AUC (0.943) among all models evaluated. Additionally, the Radexpand16_Clinica_R model exhibited excellent sensitivity, specificity, and favorable clinical utility as shown by DCA. Conclusions The decision-based fusion model Radexpand16_Clinica_R effectively differentiates DLBCL from HL. Incorporating peritumoral regions improved the predictive capability of the radiomics models.