Pre-therapeutic prediction of cachexia in lymphoma patients using [18F]FDG PET radiomics of muscle and fat.

BACKGROUND: Cachexia adversely affects treatment outcomes in patients with lymphoma, highlighting the need for early risk identification. This study aimed to develop a predictive model using [¹⁸F]fluoro-2-deoxy-D-glucose ([¹⁸F]FDG) positron emission tomography (PET) radiomics features to identify lymphoma patients at risk of developing cachexia. METHODS: A total of 150 lymphoma patients who underwent pre-treatment [¹⁸F]FDG PET/computed tomography (CT) were retrospectively enrolled from two centers and randomly divided into training and testing cohorts. Radiomics features were extracted from metabolic tissues, including the liver, visceral fat, subcutaneous fat, psoas muscle, and sacrospinal muscle. Three models were constructed: radiomics, clinical, and a combined model integrating both. Model performance was evaluated using area under curve (AUC) and AUCs were compared using DeLong's test. RESULTS: In the training cohort, 61 of 105 patients developed cachexia; in the testing cohort (n = 45), 26 developed cachexia. The radiomics model incorporated five features from subcutaneous fat, visceral fat, and psoas muscle. The combined model, incorporating radiomics and clinical features, achieved an AUC of 0.916 in the training cohort, significantly outperforming the clinical (AUC = 0.844; P = 0.013) and radiomics (AUC = 0.826; P = 0.005) models. In the testing cohort, the radiomics (AUC = 0.816; P = 0.040) and combined (AUC = 0.759; P = 0.003) models significantly outperformed the clinical model (AUC = 0.601). CONCLUSION: Radiomics features from [¹⁸F]FDG PET images of visceral fat, subcutaneous fat, and the psoas muscle may effectively identify lymphoma patients at high risk of developing cachexia.