Automated brain tumor detection using advanced deep learning models

Abstract Purpose This study explores the application of deep learning, specifically Convolutional Neural Networks (CNNs), for the detection and classification of brain tumors in medical images. The research addresses the challenges of tumor differentiation and network distribution complexity under various models of deeplearning and to find out which model is most effective. The state of the art as well as classical models —alongside two complementary models, were evaluated for their performance in classifying brain tumors. The study utilized a comprehensive dataset of brain MRI scans, representing various tumor types, sizes, and locations. Image preprocessing techniques, including normalization and scaling, were employed to enhance model performance. Each CNN model was meticulously designed, trained, and tested, with a focus on accuracy, sensitivity, specificity, and computational efficiency. Results The comparative analysis of the models revealed distinct strengths and weaknesses, with each architecture showing varying degrees of effectiveness in brain tumor classification tasks. Key findings include the models’ ability to accurately detect and differentiate between different tumor types, with specific models excelling in certain performance metrics. Conclusion This study provides insights into the suitability of various CNN models for real-world clinical applications in automated brain tumor detection. The results contribute to the ongoing advancements in medical imaging, highlighting the potential of deep learning in improving diagnostic precision and reducing the need for manual intervention and MobileNet has turned out to be best model under our observation with respect to the given data set available. MobileNet achieved 96.6% accuracy, outperforming deeper architectures such as ResNet-50 (94.8%), DenseNet-121 (95.2%), and InceptionV3 (93.7%), demonstrating the strong capability of lightweight CNNs for reliable brain-tumor classification. Beyond empirical performance, the proposed model demonstrates strong real-world applicability, offering potential integration into hospital triage systems for rapid MRI pre-screening and deployment in lightweight mobile or edge-based diagnostic tools, enabling faster and more accessible brain-tumor assessment in resource-limited settings.