Discrete Wavelet transform based Multiscale Deep CNN cascaded LSTM model for the classification of Brain Tumor

Abstract

Abstract A brain tumor denotes an anomalous mass or collection of cells that develops within the brain. We have considered about the three categories of brain tumors among the various 120 categories namely Glioma, Meningioma and Pituitary along with No Tumor. Automated classification of different brain tumor categories using Magnetic Resonance Imaging (MRI) brain scans is this paper's unique approach. In our proposed framework, we have introduced a cascade of multiscale deep Convolution Neural Network (CNN) and Long Short Term Memory (LSTM) Network to classify brain tumors using brain tumor MRI image dataset where the source images are first decomposed to extract modes from the MRI images using the Single-level Discrete 2D Wavelet Transform (dwt2) is applied. With the aim of our research, the segmentation of a single MRI image is partitioned into four modes: Contained within the available images are : a diagonal image, a vertical detail image, a horizontal detail image, and an approximation detail image. For the purpose of classifying brain tumours into Glioma, Meningioma, Pituitary, and No Tumour, the evaluated modes are fed into a multiscale deep convolution neural network (CNN) cascaded with an LSTM network. The 2000 image MRI dataset that is publically available is used to assess the proposed deep learning architecture. The outcomes demonstrate that the suggested method attained peak precision of 89.5% for Multi-Nomial classification and 98.5% for two-class classification when utilizing MRI images from the dataset. These accuracies were obtained utilising 5-Fold Cross-Validation (CV) for the Multi-Class scheme and the Hold-Out Validation method for the binary scheme.