Performance Analysis of Various Deep Learning Architectures for Detection of Meningioma from MRI Brain Images

Abstract

A brain tumor is potentially a fatal condition brought on by the unchecked growth and proliferation of brain cells. Magnetic resonance imaging (MRI) is an imaging modality used for the detection and diagnosis of brain tumors. The detection of brain tumors from images of the brain using MRI technique has been automated with the development of artificial intelligence. The development of machine learning techniques and deep learning algorithms facilitated detection with higher accuracy. Brain tumors may be either benign or malignant. Based on the specific cell type affected, brain tumors are classified into astrocytomas, oligodendrogliomas, ependymomas, medulloblastomas, pituitary tumors, and meningiomas. This research work involves deep learning architectures to detect meningioma type tumors from MRI brain images. Among the various deep learning models available, this research work focuses on LeNet and VGG 16. These models have been modified and applied for the detection and segmentation of meningioma from MRI brain images. Modified architectures of LeNet, VGG16, UNet, and CNN have been applied to detect and segment the tumor. The segmented images from UNet and modified CNN have been applied for severity detection. newlineThe main objective of the research is to differentiate a tumor affected brain image from a healthy brain image. The tumor affected image is further segmented using LeNet and VGG16 deep learning models. The deep learning model UNet has been applied to the segmented images to classify the severity of the tumor as early or severe. Further, a modified CNN has also been developed to classify the severity. The performance of the above models in segmentation techniques and severity classification has been studied. The proposed method stated is experimentally tested on the sets of meningioma and healthy brain images on both the Nanfang and Kaggle datasets. Preprocessing of brain MRI images, feature computations, and feature classification have been done through the proposed Modified LeNet model.