Abstract
BACKGROUND: To develop radiomics methods for predicting response to epidermal growth factor receptor (EGFR)-tyrosine kinase inhibitors (TKIs) based on computed tomography (CT), magnetic resonance imaging (MRI), and positron emission computed tomography (PET) from primary lesion and brain metastasis (BM) in patients with non-small cell lung cancer (NSCLC). METHODS: The retrospective study enrolled 140 patients between May 2017 and December 2022 from Center 1, and 45 patients between January 2015 and March 2024 from Center 2. All patients underwent an (18)F-FDG PET/CT scan. Patients with BM underwent brain MRI. Radiomics features were extracted from both primary lesion and BM and selected using max-relevance and min-redundancy (mRMR) method. A logistic regression model was developed based on primary lesion and BM, and evaluated using receiver operating characteristic (ROC) curve analysis, calibration, and decision curve analysis (DCA) to evaluate the applicability. RESULTS: A total of 7 and 5 important features were selected from primary lesion and BM, respectively. The SUVmax was identified as the most predictive clinical indicator. The developed nomogram based on primary lesion generated AUCs of 0.832, 0.775 and 0.798 in training, internal validation and external validation cohorts, respectively. The nomogram based on BM yielded AUCs of 0.985, 0.921 and 0.956 in training, internal validation and external validation cohorts, respectively. CONCLUSION: This study indicated that radiomics features from PET/CT and MRI of primary lesion and BM can be predictive of response to EGFR-TKI. The constructed nomogram may be considered as a non-invasive tool to stratify which patient can benefit from the EGFR-TKI therapy.