<![CDATA[Brain tumors are abnormal cell growths in brain tissue that can endanger patient safety if not detected early. Diagnosis is generally performed by radiologists through the examination of Magnetic Resonance Imaging (MRI) images; however, this approach requires considerable time and may introduce subjectivity. This study aims to compare the performance of two Convolutional Neural Network (CNN) architectures, namely ResNet50V2 and VGG16, in classifying brain tumors on MRI images, and to implement the best-performing model into a Streamlit-based application prototype. The dataset used was the Brain Tumor MRI dataset, which consists of four classes: glioma, meningioma, pituitary, and no tumor. Both models were trained using the transfer learning approach and evaluated using the metrics accuracy, precision, sensitivity, specificity, and F1-score. The results show that ResNet50V2 achieved higher performance than VGG16 across all evaluation metrics, with an accuracy of 88.17% compared with 86.19%. This advantage is likely related to the residual connection mechanism in ResNet50V2, which helps address the vanishing gradient problem, whereas VGG16 has a simpler architecture. The best-performing model was then implemented as a Streamlit-based application prototype for classifying brain tumor MRI images. However, the prototype remains at an early research stage and requires further testing using clinical data and evaluation by medical professionals before it can be considered a diagnostic support tool.]]>
