Enhancing Brain Tumor Identification with EfficientNetB4 in MRI Imaging

Brain tumors pose significant health risks, making early and accurate detection crucial. While MRI is effective for tumor identification, manual interpretation is time-consuming and error-prone. This study develops an automated system for brain tumor detection using pretrained deep learning models, including DenseNet121, AlexNet, VGG16, VGG19, ResNet50, InceptionV3,and EfficientNet variants. EfficientNetB4, in particular, achieves the highest validation accuracy. The model is further optimized through fine-tuning and additional layers, enhancing its diagnostic performance on a publicly available multiclass brain tumor dataset. The customized pretrained EfficientNetB4 model achieves impressive results, with class accuracies of 100% for no tumor, 100% for meningioma, 99.8% for pituitary tumors, and 98.7% for glioma tumors. With a test accuracy of 98.8% and a validation rate of 99.5%, the overall accuracy comes in at 99.1%. These results highlight how well our suggested solution uses modern deep learning methods to improve the efficiency and accuracy of brain cancer recognition.