MRI Radiomics Meets Artificial Intelligence: Development and Validation of a Random Forest Model for Brain Tumor Classification in Sudanese Patients

<title>Abstract</title> <p>Background/Objective: Differentiating benign from malignant brain tumors on conventional MRI remains challenging because of overlapping imaging features and inter-observer subjectivity. Radiomics combined with machine learning offers a quantitative, non-invasive means of improving diagnostic accuracy. This study aimed to develop and evaluate a predictive model for classifying brain tumor type among Sudanese patients using MRI-based radiomic features and a Random Forest algorithm. Methods A retrospective dataset of 120 histopathologically confirmed cases (60 benign, 60 malignant) was analyzed. MRI images underwent standardized preprocessing, manual tumor segmentation, and radiomic feature extraction using PyRadiomics in accordance with Image Biomarker Standardization Initiative (IBSI) recommendations. Six representative features were retained after manual feature selection. The dataset was split into training (70%) and testing (30%) sets, and a Random Forest classifier was developed with five-fold cross-validation. Model performance was evaluated using accuracy, sensitivity, specificity, F1-score, and the area under the receiver operating characteristic curve (AUC). Results The model achieved an accuracy of 71%, sensitivity of 75%, specificity of 66.7%, an F1-score of 0.72, and an AUC of 0.764, indicating good discriminative performance. No statistically significant association was found between tumor type and demographic variables (p &gt; 0.05), although a non-significant trend of increasing malignancy with age was observed. Tumor volume was significantly associated with the presence of neurological symptoms (p &lt; 0.01). Conclusion MRI-based radiomics combined with a Random Forest classifier provides a feasible, non-invasive approach for brain tumor classification. The model demonstrates practical applicability in resource-limited settings and may support clinical decision-making and diagnostic accuracy within the Sudanese healthcare system, pending validation in larger, multicenter cohorts.</p>