An optimized XGBoost technique for accurate brain tumor detection using feature selection and image segmentation

Abstract

An abnormal multiplication of cells in the brain forms malignant and benign brain tumors. Malignant brain tumors are more prevalent than benign ones. Detecting a tumor’s physical features may be tedious and time-consuming for medical experts due to the complexity of a tumor’s structure and noise growth in Magnetic Resonance Imaging (MRI) data. Detecting and localizing tumors at an early stage in their growth is crucial. Complex structures and precise clinical identification using segmentation are considered the norm in medical imaging. As the quantity of images rises, the radiologist’s manual tumor assessment might result in an incorrect conclusion. Due to the likelihood of human error, the evaluation and categorization of medical images require an automated method. This paper optimizes the eXtreme Gradient Boosting (XGBoost) approach through image processing and feature selection for the reliable identification of brain tumors. Images are improved using the Contrast-Limited Adaptive Histogram Equalization (CLAHE) method. The K-Means algorithm divides images into sections. This segmentation assists in pinpointing the specific area of interest. Particle Swarm Optimization (PSO) is used to choose characteristics. The data is classified by XGBoost, Naive Bayes, and the Iterative Dichotomiser 3 (ID3). According to the experimental results, the suggested PSO-XGBoost model achieved 97% accuracy, 97% specificity, 98% precision, and 98% recall.