Revolutionizing MRI-Based Brain Tumor Classification with BrainMRI-NetX for Superior Accuracy and Reliability

Abstract

This study aims to enhance and ensure reliable MRI-based brain tumor classification through the development of an innovative BrainMRI-NetX model, incorporating advanced techniques such as Depthwise Separable Convolutions, Residual Blocks, Squeeze-and-Excite Blocks, and Self-Attention Layers. For feature extraction, we utilized a hybrid VGG19 and LSTM model. Our primary goal is to develop and evaluate a CNN model that outperforms state-of-the-art models in terms of F-score, recall, accuracy, and precision. The proposed BrainMRI-NetX model was trained using cutting-edge optimization techniques on a large dataset of FigShare MRI brain images, significantly enhancing its performance. We thoroughly evaluated the model's critical performance indicators: F-score, recall, accuracy, and precision. When benchmarked against popular models such as ResNet-152, DenseNet121, and VGG16, our proposed model demonstrated superior performance, achieving an F-score of 0.96, and recall, accuracy, and precision all at 0.99. In comparison, DenseNet121 showed an accuracy of 0.85, precision of 0.89, recall of 0.90, and F-score of 0.88. ResNet-152 and VGG16 exhibited lower performance metrics, with accuracy at 0.86, precision at 0.85, recall at 0.84, and F-score at 0.87. The exceptional performance of our proposed BrainMRI-NetX model highlights its potential for advancing medical diagnostics, particularly in MRI-based brain tumor classification.