Experimental Analysis of Deep Learning Algorithms Used in Brain Tumor Classification

Abstract

Timely detection and analysis of brain tumors is necessary for saving lives of people in world. In recent times, deep learning using transfer learning (TL) approaches are commonly used for identifying 3 common types of tumors, e.g., meningioma, glioma, and pituitary glands. We used pre-trained transfer learning techniques to identify meningioma, glioma, and pituitary brain tumors. The goal of this research is to evaluate and compare the performance of various deep learning algorithms which can be used for brain tumor classification. Six pre-trained TL classifiers are used in the experimental analysis. InceptionV3, Xception, Resnet50, EfficienNetB0, VGG16, and MobileNet use a fine-grained classification approach to automatically identify and classify brain tumors. The experimentation is conducted on brain tumor MR image dataset with 7022 images available freely on Kaggle, and the tool used is Python. The results are computed using commonly used matrix. Classification experiments have shown that the VGG16 TL algorithm showed excellent performance and achieves the highest accuracy of 99.09% in the detection and classification of no tumors, gliomas, meningiomas, and pituitary brain tumors. Other models’ except EffecientNet B0 are also performing satisfactorily.