Abstract
Brain tumors are abnormal growths of brain cells that can affect both children and the elderly, and their prevalence has been increasing in today's world. Recent data from the World Health Organization (WHO) indicates that Asia has the highest number of brain tumor diagnoses, which poses a life-threatening risk. However, detecting tumors is challenging because of the heterogeneity of tumor cells. Fortunately, the application of Deep Learning (DL) and Machine Learning (ML) highly facilitates the diagnosis of brain tumors using Magnetic Resonance Imaging (MRI). Pre-trained Convolutional Neural Network (CNN) models, a type of Deep Neural Network (DNN), combined with ML models, could be used to detect the presence of a tumor in MRI images. In this work, to detect or classify brain tumors, we have employed CNN models including DenseNet121, VGG19, InceptionV3, InceptionResNetV2, and EfficientNetB0, together with classifiers such as a custom classifier, Support Vector Machine (SVM), XGBoost, and Random Forest (RF). The dataset incorporated in this work includes four categories of MRI images, i.e., glioma, meningioma, pituitary tumors (considered as the tumor cases), and MRI images without these (considered as the non-tumor cases). From this brain tumor dataset, 3,264 images are utilized, with 80\% allocated for the training and 20\% for testing. The images are resized to 244×244 pixels, normalized within the range of [0, 1], and transformed with the Gabor filter to improve overall performance. Our classification approach is evaluated with accuracy, precision, recall, F1 score, and AUC values, with necessary graphs. Our DL-based custom classifier with the DenseNet121 achieves an impressive classification accuracy of 99.82%.
