Lesion Localization and Extreme Gradient Boosting Characterization with Brain Tumor MRI Images

Abstract

Detection of tumor lesion with their precise location and characterization is an important task in the brain tumor diagnosis from the magnetic resonance (MR) images. But it is a time-taking task and error-prone process by radiologists or clinical experts. Several works have been introduced in brain tumor detection but it failed to discover the exact location and characterization of the lesion. In order to improve the automatic tumor lesion localization and characterization, an efficient machine learning technique called Lee Filtered Bivariate Correlative Regression based Extreme Gradient Boosting (LFBCR-EGB) is introduced. Initially “n” numbers of MRI brain images are taken from the database. The LFBCR-EGB technique comprises three major processes, namely preprocessing, lesion localization, and characterization. The regression function is used to find the positive and negative similarity between the pixels in an image. The negative similarity result provides the exact localization results with minimum time. Finally, the lesion characterization is done by applying an extreme gradient boosting technique to improve the accuracy with certain features. The features are extracted from the ROI and they construct several weak learners. The ID3 classifier is used as weak learners of the extreme gradient boosting technique to classify normal or abnormal tissue based on the information gain. In this way, the LFBCR-EGB technique performs accurate and fast tumor detection along with the exact location of the tumor. Experimental evaluation of the proposed LFBCR-EGB technique is carried out using MRI brain image database with different factors such as peak signal-to-noise ratio, lesion localization time, and classification accuracy. The simulation results show that the proposed LFBCR-EGB technique obtains better results in terms of peak signal ratio and classification accuracy with lesion localization time.