Machine Learning on FDG PET/CT Predicts Lymphoma Bone Marrow Infiltration

A random forest model using clinical, lab, and FDG PET/CT data predicted bone marrow infiltration in lymphoma with an AUC of 0.842 and accuracy of 80.9% in a retrospective testing set.

• Key predictors included platelet count, hemoglobin, LDH, lymph node involvement, bone infiltration characteristics, and SUVmax of sternum, thoracic/lumbar vertebrae, sacrum, and femur.
• In the pathology-confirmed subgroup, the RF model maintained performance (AUC 0.846, accuracy 82.5%).

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.