Glioma/Glioblastoma Detection in Brain MRI using Pre-trained Deep-Learning Scheme
V. Rajinikanth, Seifedine Kadry, Robertas Damaševičius, R Sujitha, Gangadharam Balaji, Mazin Abed Mohammed
Abstract
Convolutional Neural Network (CNN) supported medicinal image examination is extensively accepted due to its reputation and improved accuracy. The investigational outcome obtained with DLS along with a chosen classifier helps to achieve better detection results than the traditional and machine-learning methods. The proposed research examines the performance of pre-trained VGG16 and VGG19 schemes in detecting the brain tumour (Glioma/Glioblastoma) grade using different pooling methods. The classification is performed using SoftMax with five-fold cross-validation, and the products are compared and presented. The brain tumour images considered in this study are collected from The Cancer Imaging Archive (TCIA) dataset. This work considered 2000 images (1000 Glioma and 1000 Glioblastoma) of axial-plane with dimension of 224×224×3 pixels for the assessment, and the attained results are compared. The experimental outcome achieved with Python® confirms that the VGG16 with average-pooling provides a better classification accuracy (>99%) with Decision Tree (DT) compared with other methods considered.