Abstract
PURPOSE: Lung cancer is one of the most common types of cancer, and the presence of brain metastases has a significant impact on the clinical course and prognosis. EGFR, BRAF, ALK, and ROS1 mutations have previously been identified in lung cancer, and knowing the tumour mutation status is important for molecular therapy. In our study, we investigated the performance of radiomics in predicting the status of brain metastases detected by brain magnetic resonance imaging (MRI), a noninvasive method, in with brain metastases patients diagnosed with lung cancer. MATERIAL AND METHODS: Lung cancer cases with brain metastasis in our hospital between 2014 and 2024 were analysed retrospectively. Histopathological data were obtained from tissue biopsy results, and EGFR, BRAF, ALK, and ROS1 mutation status were recorded. A total of N = 84 patients were included in the study, and 107 original radiomics parameters were obtained from the segmentation files extracted from the patient images. Due to the class unbalance, the performance of the model was tested using the stratified folding method. RESULTS: Five (6.02%) of the patients had EGFR, 3 (4.17%) had ALK, and 2 (2.78%) had ROS1 mutations. Model 1 used for EGFR mutation prediction showed high performance with 93.82% accuracy, Model 2 used for ALK with 84.76% accuracy, and Model 3 used for ROS1 with 79.33% accuracy. CONCLUSION: Our study showed that EGFR mutations, in particular, can be detected with high accuracy by radiomics in lung cancer patients with brain metastases without additional invasive procedures.