Smart architectures: computerized classification of brain tumors from MRI images utilizing deep learning approaches

Brain tumors are complex medical conditions that appear due to the abnormal proliferation of cells. They are tumors (benign or malignant) that grow abnormally or are masses of tissue that develop in a dreadful manner in the human brain. In most cases, there is no specific or known cause for the occurrence of these tumors, and late diagnosis leads to a disproportionate progression of cancer cells or the growth of malignant masses in the brain, which leads to difficulty in performing treatment directly. Deep learning methods, most notably convolutional neural networks (CNNs), are employed to accurately diagnose and classify brain tumor images due to their ability to process large sets of complex medical images. In this paper, the role of deep learning approaches will be highlighted by utilizing different architectures of the CNNs (DenseNet-201, Inception-V1, AlexNet, and MobileNetV2), comparing their performance, and finding the most suitable solution for identifying and classifying brain tumors and solving the problem of late or incorrect diagnosis with high efficiency and accuracy. The database of this paper includes more than 3000 MRI images taken from Kaggle. During implementation, it became clear that the most suitable performance was the MobileNetV2, as it earned an accuracy of 96.5%, a sensitivity of 96.6%, a precision of 98.2%, a specificity of 96.3% and an F1-score of 97.4%. These effects demonstrate that MobileNetV2 is the best architecture among others, as it is highly capable of classifying aberrant and normal tumors from brain MRI images. This paper emphasizes that deep learning is crucial in supporting hospitals and experts in interpreting MRI images.