An enhanced performance analysis of mri images for early detection of brain tumor using deep learning techniques

Abstract

Automatic segmentation and classification of medical images acts as a major part in growth forecast, diagnostics and the handling of brain tumors. Thus, early detection of brain tumor assists in better treatment and increases the survival rate of patients. However, manual procedures have been conducted on a large-scale clinical image databases in medicinal practices that take more time and cost. Hence, there exists a requirement of the automated brain tumor detection model. It is developed by adopting intelligent tools for both segmentation and classification approaches. The automated brain tumor detection consists of several stages like pre-processing, tumor segmentation, feature extraction and classification. In the first phase, the model has been divided into four phases, in which the initial phase gathers the MRI images from benchmark sources, which is forwarded to the pre-processing stage that is performed through sequential processes like Image scaling, entropy-based trilateral filtering, and skull stripping. Then, Fuzzy Centroid-based Region Growing Algorithm is employed for segmenting the tumor, where the segmented images undergo the feature extraction phase through the first order and the higher order statistical measures. A hybrid classifier with the integrated Convolutional Neural Network (CNN) and Neural Network (NN) is introduced in this initial stage, where NN takes input as the first order and the higher order statistical measures while the CNN considers input as the third level DWT image. Here, a novel technique Cross Over Rooster - Chicken Swarm Optimization (COR-CSO) algorithm is adopted by tuning the hidden neuron count of both CNN and NN. Finally, the AND operation is carried out for getting the classified outcomes as normal and abnormal newline