Detection of Brain Tumors in MRI Scans Utilizing Deep Learning: A Comparative Analysis of Advanced CNN Models

Abstract

Classification of brain tumors is essential in medical diagnostics, as timely and precise identification can significantly enhance patient outcomes. This study examines the efficacy of pre-trained deep learning models in classifying brain MRI images into four categories: pituitary, meningioma, glioma, and no tumor, with the aim of automating and enhancing the diagnostic process. We utilized a publically accessible MRI dataset comprising 7,023 pictures of brain tumors. Alongside comprehensive image preprocessing and data augmentation, transfer learning was employed to enhance the performance of four sophisticated convolutionalneural network architectures: DenseNet121, ResNet50, Xception, and MobileNet. The models were successfully refined by transfer learning, reducing computational requirements while enhancing classification precision. DenseNet121 surpassed Xception, MobileNet, and ResNet50 in the evaluated models, attaining the greatest accuracy of 98.47% and an F1 score of 98.47%. The models’ appropriateness for clinical application was validated by their robust generalization and consistent performance across critical assessment parameters. To enhance memory for particular tumor classifications and render deep learning predictions more interpretable in medical settings, further developments are required.