Revisiting Multi-Nodal Radiomics with Advanced Feature Learning for Lymphoma Classification

A multi-center machine learning framework for patient-level classification of Hodgkin and non-Hodgkin lymphoma from baseline FDG-PET imaging.

The project represents patients as variable-size sets of involved lymph nodes and investigates advanced aggregation methods, including Transformers, Deep Sets, multiple instance learning, Set Transformers, and graph-based models. It also explores test-time adaptation to improve model robustness and generalization across clinical centers.