Brain tumors, which are abnormal cell growths that deviate from the typical brain tissue development, pose a serious and potentially life-threatening condition. Early detection and classification of these tumors are critical for providing the most effective treatment. Magnetic Resonance Imaging (MRI) is a widely used diagnostic tool for brain tumors due to its ability to produce high-resolution images of soft tissues. However, manual analysis of MRI images is time-consuming, subjective, and prone to human error. Machine learning algorithms, particularly deep learning models, have shown great potential in automating this task with high precision. The objective of the proposed method is to first detect the presence of a tumor and then classify it into one of three major types: meningioma, glioma, or pituitary tumor. The approach presented in this paper involves using ENet for segmentation. Afterward, feature extraction is performed using the Gray Level Co-occurrence Matrix (GLCM). For classification, the method employs an ensemble of pre-trained models, including ResNeXt-50, DenseNet-169, and Inception V3 for binary classification (benign vs. malignant) and ShuffleNet V2, DenseNet-169, and MnasNet for multi-class classification (glioma, meningioma, and pituitary). This hybrid algorithm will be assessed using a dataset of brain cancer images, with the goal of achieving improved accuracy in both tumor detection and classification by tumor type and location.
Kothandan et al. (Tue,) studied this question.