Exploring the Role of Peritumoral Edema in Predicting Lung Cancer Subtypes Through T2-FLAIR Digital Subtraction Imaging of Metastatic Brain Tumors

Abstract

Background/Objectives: This study aimed to investigate whether small-cell lung cancer (SCLC) and non-small-cell lung cancer (NSCLC) can be distinguished based on radiomics data derived from T2-FLAIR digital subtraction images of the peritumoral edema region in patients with brain metastases. Methods: A total of 136 patients who underwent surgery for brain tumors, including 100 patients in the Pretreat-Metstobrain-MASKS dataset and 36 patients from our institution, were included in our study. Radiomic features were extracted from digitally subtracted T2-FLAIR images in the peritumoral edema area. Patients were divided into NSCLC and SCLC groups. The maximum relevance-minimum redundancy (mRMR) method was then used for dimensionality reduction. The Naive Bayes algorithm was used for model development, and the interpretability of the model was explored using SHapley Additive exPlanations (SHAP). The performance metrics included the area under the curve (AUC), sensitivity (SENS), and specificity (SPEC). Results: The mean age of NSCLC patients was 64.6 +/- 10.3 years, and that of SCLC patients was 63.4 +/- 11.7 years. In the external validation cohort, the model achieved an AUC of 0.82 (0.68-0.97), a SENS of 0.87 (0.74-0.91), and a SPEC of 0.72 (0.72-0.89). In the train cohort, the model achieved an AUC of 1.000, a SENS of 1.000, and a SPEC of 1.000. The feature providing the best effect was wavelet-HHHglcmJointEnergy, with a SHAP value of approximately 2.5. Conclusions: An artificial intelligence model developed using radiomics data from T2-FLAIR digital subtraction images of the peritumoral edema area can identify the histologic type of lung cancer in patients with associated brain metastases.