An interpretable PET/CT-based radiomic-clinical model for predicting bone marrow involvement in follicular lymphoma: comparison of pelvic and spine-pelvis VOI frameworks - PubMed
a day ago
- #Radiomics
- #PET/CT
- #Follicular Lymphoma
- This retrospective study developed an interpretable PET/CT-based radiomic-clinical model using baseline 18F-FDG PET/CT to non-invasively predict bone marrow involvement (BMI) in follicular lymphoma (FL).
- Two bone marrow volume-of-interest (VOI) frameworks were compared: pelvic and spine-pelvis, with the spine-pelvis VOI framework, combined with clinical, conventional imaging, and radiomic features, producing the best-performing model.
- The gradient boosting machine (GBM) model within the optimal scheme achieved the highest performance on the independent validation set (AUC=0.906, Accuracy=0.877, Precision=0.926, Sensitivity=0.833, Specificity=0.926, F1 score=0.877).
- BMI in FL was associated with higher tumor burden and altered metabolic heterogeneity in bone marrow, indicated by features like widespread regional lymph node involvement, B symptoms, and abnormal levels of Hb, LDH, Apo A, eGFR, and β2-MG.
- SHAP analysis highlighted key contributors to the model, including LNr, β2-MG, PET-Orig-FO-IQR, and various texture features from wavelet/LBP transformations, suggesting imaging phenotypes involve increased metabolism and spatial heterogeneity in bone marrow.