<![CDATA[Automatic classification of brain tumors from MRI images is crucial for supporting early diagnosis and improving treatment planning. However, manual diagnostic processes remain limited by subjectivity and resource constraints. This study aims to optimize brain tumor classification by conducting a comparative analysis of six Convolutional Neural Network (CNN) architectures: VGG16, VGG19, MobileNet, InceptionV3, AlexNet, and Xception. The MRI datasets were sourced from open repositories and processed through normalization, noise reduction, segmentation, and data augmentation. All CNN models were implemented using transfer learning and trained under consistent parameters. Model performance was evaluated based on accuracy, sensitivity, specificity, and F1-score. The results revealed that the Xception and InceptionV3 architectures achieved the highest classification performance, with validation accuracies of 97.9% and 96.1%, respectively. MobileNet also performed competitively at 95.6%, offering notable computational efficiency. In contrast, VGG19 and AlexNet yielded lower validation accuracies and exhibited signs of overfitting. These findings highlight the effectiveness of modern CNN architectures that incorporate depthwise separable convolutions and residual connections in extracting complex features from brain MRI images. Therefore, models such as Xception and MobileNet are strong candidates for implementation in computer-aided diagnosis systems in resource-constrained clinical environments.]]>
Copyrights © 2025
