<![CDATA[Brain tumors are neurological conditions that affect essential bodily functioning, necessitating a prompt and precise diagnosis. Brain tumors are frequently detected with magnetic resonance imaging (MRI), but manual interpretation is laborious and heavily reliant on radiologists' skill. Deep learning has become a popular method for improving medical image analysis. This paper suggests a hybrid deep learning architecture for MRI image-based brain tumor identification that combines MobileNetV3 and U-Net. U-Net was utilized for tumor region segmentation, and MobileNetV3 was utilized for tumor categorization. A publicly available Kaggle Brain Tumor MRI dataset and the BraTS 2021 Dataset were used in the study. The accuracy, precision, recall, F1-score, confusion matrix, AUC-ROC, dice coefficient, and intersection over union (IoU) metrics were used to assess the model's performance. The MobileNetV3 model exhibited good precision, recall, and F1-score values along with 95% classification accuracy. Furthermore, outstanding classification performance was demonstrated by the AUC-ROC values, which reached 0.99 and 1.00. The U-Net model received a Dice Coefficient of 0.9404, an IoU score of 0.8940, and a validation accuracy of 0.9936 for segmentation. For precise brain tumor detection and visualization on MRI images, the suggested hybrid architecture successfully integrates classification and segmentation tasks.]]>
Copyrights © 2026
