非小細胞肺癌患者における潜在的リンパ節転移を予測するFDG-PETベースのレジオニクスモデルの開発と外部検証

Accurate detection of occult lymph node metastasis (OLNM) in patients with localized non-small cell lung cancer (NSCLC) remains a clinical challenge. This study aimed to develop and validate a radiomics-based predictive model for OLNM.

A radiomics model (Model_PET) and a model (Model_Combined) combining radiomics and clinical features were developed using a retrospective monocentric cohort of localized NSCLC patients treated with surgery (Cohort A) and tested on an external cohort (Cohort B) of 112 localized NSCLC patients also treated with surgery (publicly available Radiogenomics cohort). The model was further assessed in an independent cohort of 488 patients with localized NSCLC who underwent definitive stereotactic body radiotherapy (SBRT) (Cohort C) using regional relapse free survival (RRFS) as a surrogate for OLNM. Radiomic features were extracted from pre-treatment FDG PET and combined to predict OLNM using a multilayer perceptron approach.

In the training cohort, the Model_PET and Model_Combined achieved AUCs of 0.92/0.99 and balanced accuracies (Bacc) of 80.0%/85.3%, respectively. In the Cohort B, the Model_PET and Model_Combined resulted in AUCs of 0.73/0.67 and Baccs of 71.2%/51.7%, respectively. In the Cohort C, the predicted OLNM risk based on Model_PET was significantly associated with worse RFFS (HR 1.60 95% CI 1.03-2.48, p = 0.04). The Model_Combined was not associated with survival outcomes (p > 0.05).

This study presents a radiomics-based predictive model for OLNM in localized NSCLC, validated across several retrospective independent cohorts. Subject to a prospective evaluation, the model could be used to refine clinical decision-making.