Brain Tumor Detection and Classification with One-Hot Encoding and EfficientNetB0 using MRI Images

Convolutional Neural Networks (CNNs) perform well in accurately classifying brain tumors identified in medical scans such as MRI. This study presents a CNN architecture, which contains convolutional layers for extracting features followed by maximum pooling layers for spatial down-sampling, for dimensionality reduction. To mitigate overfitting, dropout layers are employed are strategically integrated, ensuring the generalizability within the model. The task is accomplished, incorporating using fully connected layers with the SoftMax activation function. The Convolutional Neural Network(CNN) proposed architecture demonstrates effectiveness in categorizing brain tumors into three distinct types: meningioma, glioma, and pituitary tumors. Experimental evaluation reveals promising results, with the model achieving an overall classification accuracy of 98%. Specifically, it detects glioma with 96% accuracy, identifies no tumor with 99% accuracy, differentiates meningioma with 97% accuracy, and identifies pituitary tumors with 99% accuracy. The dataset comprises 3264 images, 90% of which are for training and 10% for testing. The approach shows considerable promise to assist clinicians in accurate and timely diagnosis, thereby facilitating tailored treatment planning for patients with brain tumors. Further research can explore improvements to the network architecture and explore its applicability in different medical imaging datasets. © 2024 IEEE.