A new clinical diagnosis system for detecting brain tumor using integrated ResNet_Stacking with XGBoost

Abstract

The cancer disease prediction and detection processes are crucial tasks in this emerging world and it is tough to manage the diseases. Generally, the disease prediction processes are done by using various kinds of inputs including numerical data and Magnetic Resonance Imaging (MRI) images. In the past, the researchers use the segmentation and classification methods on MRI images to enrich performance of the disease prediction system. However, these input images are using for the prediction, detection and diagnosis with reasonable delay and less prediction accuracy. In this work, we propose a new brain tumor disease prediction and diagnosis system that incorporates the segmentation and classification techniques for predicting and detecting the cancer diseases effectively. Here, new data pre-processing methods like mean fusion, automated cropping of the Region of Interest (ROI), and Gaussian filtering are used for performing effective classification. Moreover, the proposed system performs the standardized resizing and rescaling processes over the MRI images and it also automate the cropping process of the ROI to ensure consistent input sizes. In addition, the brain tumor segmentation process is performed by applying the LinkNet architecture with a SEResNeXt101 backbone network. Finally, the system focuses on brain tumor classification using ensemble model which combines the ResNet architectures through stacking, leveraging the strengths of individual models to improve classification accuracy in the first stage, and it incorporates the XGBoost algorithm to enhance the performance further in efficient manner. The proposed disease prediction system is assessed through experiments on MRI images. Finally, the system is proved as superior to other available disease prediction systems with respect to sensitivity, specificity, AUC and obtained 95.84% as overall prediction accuracy.