Brain Tumor Segmentation & Classification using Optimized k-means (SFLA) and Ensemble Learning

Abstract

Brain tumor is a common disease that can occur at any age in humans. Early-stage brain tumor segmentation and classification from low-contrast MRI images is always difficult. In this paper, a new hybrid optimized k-means algorithm based on the shuffled frog leap algorithm (SFLA) followed by thresholding and morphological with ensemble learning is developed. The proposed work is divided into two segments. After pre-processing of low-contrast MRI images the brain tumor area is calculated from the segmented MRI image then the most efficient features are also extracted using discrete wavelet transformation (DWT) techniques. In the second segment, these extracted features are fed as input parameters into a trained brain tumor classifier using an ensemble learning approach. The ensemble-leaning approach model is trained by a feature dataset collected from an online source. The KNN, decision tree, gradient boosting, random forest, and ANN classifiers are used to classify the type of tumor (benign or malignant) from the low contrast brain tumor MRI image. The proposed framework is more efficient and has an accuracy (average of all models accuracy) of 98.07 percent, sensitivity of 98.21 percent, and specificity of 97.25 percent in predicting the type of brain tumor.