Differential diagnosis of glioblastom and soliter brain metastasis:the success of artificial intelligence models created with radiomics data obtained by automatic segmentation from conventional MRİ sequences

Purpose: We aimed to distinguish glioblastoma (GBM) from solitary brain metastasis with machine learning models developed with radiomics data obtained by artificial intelligence-based automatic tumor segmentation over conventional MR images of patients. Materials and Methods: Our study was conducted in a single center and retrospective at Afyonkarahisar Health Sciences University Health Application and Research Center Hospital. In the study, patients with proven diagnosis of glioblastoma and brain metastasis in the pathology records of our hospital between January 2011 and June 2020 were retrospectively screened and medical records and images in PACS system were examined. We included patients who had undergone pre-operative contrast-enhanced brain MRI and pathologic proven diagnosis. We excluded patients whose image quality was insufficient and whose pathology report was inappropriate. 35 GBM and 25 solitary brain metastasis patients were included in the study. T1-weighted, post-contrast T1-weighted, T2-weighted and T2 fluid attenuated inversion recovery (FLAIR) weighted anonymized images of the patients were uploaded to BraTumIA program. With the program, the lesions of the patients were divided into 4 different segments by artificial intelligence as necrosis, non-enhancing solid area, enhancing solid area and peritumoral edema. Subsequently, the segmentation failings were corrected manually. We have extracted radiomics features from T1 post-contrast and T2 AG FLAIR images with the "Radiomics" plug-in of 3DSlicer package program. Orange data mining program and Python Sci-Kit Learn library were used to develop artificial intelligence models. We applied a dimension reduction process to avoid overfitting and colinearity. Subsequently, Deep Neural Networks (DNN), Support Vector Machine (SVM), Random Forest (RF), Naive Bayes (NB) were modeled using the 10-fold cross-validation method with these features. We used the accuracy, sensitivity, specificity, and area under the curve (AUC) parameters to evaluate the model performance. Results: There was no difference between GBM and metastasis groups in terms of age, gender and localization. Among the machine models developed, the most successful results were obtained in neural network and support vector machine algorithms. In the neural network classifier; AUC, accuracy, F measure, positive predictive value, sensitivity, and specificity in distinguishing GBM and metastasis were 0.975, 0.917, 0.917, 0.922, 0.917, and 0.900, respectively. In the support vector machine classifier; AUC, accuracy, F measure, positive predictive value, sensitivity, and specificity in distinguishing GBM from metastasis were 0.974, 0.901, 0.900, 0.919, 0.9900, 0.929, respectively. Conclusion: In the differential diagnosis of GBM and solitary brain metastases, radiomics-based artificial intelligence models can be distinguished with high accuracy with only conventional sequences without device dependency. Key Words: radiomics, machine learning, glioblastoma, metastatic brain tumor, texture analysis, automatic segmentation