Comparative Analysis on Multiclass Brain Tumor Classification Using Deep Learning Architectures

Abnormal growth of cells inside the skull of humans, referred to as brain tumors is time that grows within the brain or surrounds the tissue. Their presence in the brain always affects human health, though they might be benign (noncancerous) or malignant (cancerous). According to data recently provided by the WHO, brain tumors account for around 2% of all cancer diagnoses, making them a serious public health problem. Brain tumor patients frequently have a variety of symptoms, such as headaches, seizures, cognitive decline, and neurological deficiencies. In this research, we have proposed a CNN model that can classify four tumor classes i.e. glioma, meningioma, pituitary, and no tumor. The images in the dataset are grayscale, and in the preprocessing step, the dimensions of the images have been changed. Parameters like batch size, epoch, optimizer, etc., are selected properly in different approaches, and finally, the data with VGG-19, Xception, and CNN architecture is from scratch. We have found accuracies (in percent) of 97.40 (VGG19), 99.24 (Xception), and 99.00 (proposed CNN model).