Brain Image Classification by Deep Neural Network with Pyramid Design of Inception Module

Abstract

In Deep Neural Network (DNN), the convolution filters of different sizes are stacked to abstract features from the input data, making weight updating difficulties and causes overfitting. A conventional Inception Module (IM) overcomes the abovementioned drawbacks by designing the architecture wider rather than deeper. In this study, DNN with Pyramid Design of Inception Module (PDIM) is designed for brain image classification using Magnetic Resonance Imaging (MRI). Depth of the architecture is increased by stacking PDIM units, and their performances are evaluated. This study's MRI images are obtained from the REpository of Molecular BRAin Neoplasia DaTa (REMBRANDT) database. The newly proposed architecture achieves 99% accuracy with 98% and 100% sensitivity and specificity. The performance comparison shows the system's effectiveness that could help the physicians for accurate brain cancer classification.