PET/CT imaging and clinical information (age, gender, smoking history, family history of cancer, history of present illness, and several tumor biomarkers, etc.) were used to establish a hierarchical multi-modal AI framework for pathological and genetic subtyping of lung cancer

The multi-modal AI framework is developed to facilitate a hierarchical and precise stratification process. The first level involves the accurate differentiation between small cell lung cancer and non-small cell lung cancer (NSCLC) in patients diagnosed with lung cancer. The second level entails the further categorization of NSCLC patients into adenocarcinoma, squamous cell carcinoma, and other less prevalent subtypes. The third level involves predicting the mutation status of the EGFR driver gene, which is most-commonly observed in patients with lung adenocarcinoma. The whole cohort was divided into the training cohort (retrospective), validation cohort (retrospective), test cohort (retrospective), and prospective cohort.

Inclusion Criteria:

• Newly diagnosed NSCLC confirmed pathologically
• Age ≥18 y
• Underwent pre-treatment 18F-FDG PET/CT scan
• No prior anti-tumor treatments
• No history of other malignancies

Exclusion Criteria:

▪ Pure ground-glass nodules with no FDG uptake

The target population of this study is patients with lung cancer who have undergone pre-treatment 18F-FDG PET/CT. The study will assess the relationships between PET/CT-based imaging features and clinical variables with the pathological subtypes.