Purpose <p>Extranodal involvement serves as a significant prognostic factor in Hodgkin lymphoma (HL), frequently correlating with adverse outcomes. Radiomics offers a data-driven approach to quantify tumor and systemic characteristics from PET/CT imaging, thereby enhancing the prediction of treatment responses. This study aims to evaluate the added value of integrating extranodal PET/CT radiomics with lesion radiomics features for machine learning (ML) -based HL early treatment response prediction, enhancing risk stratification.</p> Methods <p>A retrospective study was conducted with 110 HL patients, encompassing 420 lesions. Radiomics features were extracted from pre-treatment PET/CT scans of the lesions and five extranodal organs (lung, liver, spleen, stomach, and colon). Feature selection was performed using the minimum redundancy maximum relevance (mRMR) method. Four ML models, AdaBoost, Random Forest, XGBoost, and SVMs, were utilized to predict early treatment response. Clinical utility was assessed through decision curve analysis (DCA), and feature importance was interpreted using SHapley Additive exPlanations (SHAP).</p> Results <p>The XGBoost model demonstrated the highest performance, improving AUC-ROC from 0.969 (lesion-only) to 0.984 (combined lesion-extranodal). DCA indicated that the combined model offered greater net benefits across clinically relevant thresholds, signifying enhanced clinical utility. SHAP analysis identified that extranodal features, including spleen morphology and lung/liver texture, contributed to the model’s predictive power.</p> Conclusion <p>The integration of extranodal radiomics improves early treatment response prediction in HL. Our model demonstrates enhanced clinical utility, facilitating personalized treatment planning and better patient outcomes. This study highlights the critical role of extranodal radiomics in managing HL and supports its integration into clinical practice.</p> Clinical Trial Number <p>Not Applicable.</p> Trial Registration <p>Not Applicable.</p>