Advanced Brain Tumor Classification: A Comparative Study of CNN, EfficientNetB3, and XceptionNet Architectures

Brain tumor-which is a segment of a severe health problem-is one of the most heterogeneous pathologies in whose tumor structure the complexity itself hinders the diagnosis and treatment. The aim of this study is to evaluate the performance of several deep learning models, namely CNN, EfficientNetB3, and XceptionNet, on brain tumor type classification such as Glioma, Meningioma, Notumor, and Pituitary. In the current study, 7,023 MRI scans were utilized that were publicly available, preprocessed, and divided into training, validation, and test sets. All of them are fine-tuned and trained for optimum performance, where the performance metrics of accuracy, precision, recall, and F1-score were adapted to ascertain their efficiency. EfficientNetB3 topped the ranking list among all the used models, where it attained an accuracy of 98% in all tumor classes; hence, this model proved to be reliable. XceptionNet and CNN did quite well, where both achieved 97% and 95%, respectively, but both suffer in accurately classifying Meningioma cases. These findings then suggest that EfficientNetB3 is particularly predisposed to brain tumor classification. This study shall, therefore, bring forth the value of a number of deep learning models with large dimensions for enabling accurate and timely detection of brain tumors to improve outcomes for patients. Future studies may seek to explore the integration of such tools with explainable AI for further improvements in clinical applications.