Automated Detection and Classification Method of Brain Tumors Using CNNs and Deep Transfer Learning

Brain tumors are a challenging medical diagnostic problem with their complex morphology and limited applicability of classical detection methods. This study provides an automatic system for detecting brain tumors and classify them using Convolutional Neural Networks (CNNs). The classification and detection approaches are developed using an open-source MRI-based Brain Tumor Classification dataset comprising 6,400 scans, that have four classes: Meningioma Tumor, Pituitary Tumor, Glioma Tumor and No Tumor. For improved image quality, several augmentation techniques like rescale, rotate, shift, zoom etc. are applied and the dataset is custom balanced for better result. Three CNN models DenseNet169, EfficientNetB3, VGG16 —are analyzed for comparison on their classification. Among them, DenseNet169 and EfficientNetB3 both has achieved the maximum accuracy of 99.58%, demonstrating improved performance in distinguishing the types of tumors. Another model VGG16 achieved accuracy of 99.41%. Implementing ensemble on three models this study achieved a promising and remarkable accuracy of 99.75%. Models are also compared by some metrics like- confusion matrix, F1-score, and ROC curve, measuring models’ accuracy and effectiveness. Comparison is made with alternative approaches, including that our approach outperforms conventional machine learning methods and deep learning frameworks with a stable and efficient solution towards brain tumor classification. This paper demonstrates the potential of CNN in medical imaging for improved diagnostic efficacy and patient benefit.