Radiomic signature: A novel magnetic resonance imaging-based prognostic biomarker in patients with brainstem cavernous malformation

OBJECTIVE: Based on anatomical magnetic resonance imaging (MRI) sequences, we developed a radiomic signature for brainstem cavernous malformation patients (BSCMs) using radiomic analysis and explore its effectiveness as a prognostic biomarker. METHODS: One hundred and fourteen BSCMs with clinical, and radiomic information were collected and randomly divided into training (n = 68) and validation set (n = 46). Clinical and radiomic nomogram were constructed for the prognosis. Radiomic features were screened with three algorithms (univariate analysis, Pearson analysis, and elastic net algorithm). Cox regression model was used to build the radiomics nomogram. Finally, concordance index (C-index), time-independent receiver operating characteristic (ROC) analysis, and Decision curve analysis (DCA) were utilized to evaluate the clinical application of the radiomics nomogram. RESULTS: The radiomic signature score was calculated with 11 hemorrhage-free survival (HFS) related radiomic features from the training cohort. The patients were divided into high-risk group and low-risk group with the help of radiomic signature and the low-risk group has a better HFS than the high-risk group. In addition, three clinical characteristics including the number of hemorrhages, size, mRS, and radiomics score (Rad-score) were used to develop the radiomics nomogram. The calibration plots showed that the nomogram has good agreement between the predicted and actual survival probabilities. And, the C-index was 0.784 and 0.787 in the training cohort and validation cohort in predicting HFS; the area under curve (AUC) was 72.51 and 76.41 in the training cohort and validation cohort in 3-year survival and 67.62 and 72.57 in 5-year survival. Lastly, the DCA curve showed that the radiomics nomogram has a better clinical application than the clinical model. CONCLUSIONS: Radiomics nomogram integrating radiomics signature and clinical information showed great performance and high sensitiveness in prediction HFS in BSCMs than the clinical model.