Deep Learning Towards Multi-Classification of Tumors in Brain

Abstract

Brain tumors are recognized as one of the most deadly malformations because of its persistent effects on the brain and consequent impact on the patient's overall wellbeing. Early detection of the brain tumors is prerequisite in order to take proper treatment in due time to safeguard the valuable life of a patient. Classification and segmentation are essential for examining tumors and preferring treatments based on the types and the shape and the size of the brain tumors. Magnetic resonance Imaging (MRI) is used because of its superior quality from which the shape, size, structure, types and soft tissues of brain tumors can be easily determined. In recent times, deep learning models for recognizing brain tumors have earned a considerable interest. As a result, the CNN architecture has received the greatest deployment out of all of these deep learning models due to its extensive capabilities and adaptability. In our work, we utilized the pretrained VGG19 architecture for the classification of tumors in brain. The Unet architecture which is based on CNN is considered to be specially created for segmenting medical images.For determining the structure, shape and size of the brain tumor. Two distinct datasets have been employed for classification and segmentation tasks during the training and testing of both models. The segmentation dataset contains MRI pictures with LGG(low grade glioma), whereas the classification dataset contains four categories of brain MRI images including meningioma, glioma, no tumor and pituitary tumor. The classification architecture generates an accuracy of 92.8% and the segmentation model creates the predicted mask of the corresponding four forms of brain MRI images