Automated system for Brain tumour detection and classification using eXtreme Gradient Boosted decision trees

Abstract

This Brain tumor detection and classification is an intrinsic part of any diagnostic system and has witnessed extensive research and procedural advancement over time. The complexity of brain as an organ features to be identified, the presence of noise, poor contrast and intensity inhomogeneity in the images, efficient feature extraction, and accurate classification necessitates the development of an efficacious automated system. We propose a novel automated approach for detection and classification, using the Modified K-Means Clustering algorithm with Mean Shift Segmentation for pre-processing magnetic resonance images (MRI). Detection is done using Marker-Controlled Watershed Transform, and Gray-Level Co-Occurrence Matrix (GLCM) is used for feature extraction. For classification, we use the new and improved version of Gradient Boosted Machines (GBM) called eXtreme GBMs. Implemented using the XGBoost library, this supervised learning model has shown more accurate results and in lesser times, it is being used widely by data scientists and gives state of the art solutions.