Brain Tumor Types Classification using K-means Clustering and ANN Approach
Angona Biswas, Md. Saiful Islam
Abstract
Brain tumor is a critical disease that can cause death. For early treatment, brain tumor detection is foremost work and automatic image classification can play a vital role. Locating brain tumor, increasing accuracy level for more tumor types classification are main challenges. This paper emphasizes on proper construction of a more precise, three-class brain tumor classifier from MRI images and pays attention to remove these limitations effectively using a hybrid solution. Firstly, images are preprocessed by resizing, sharpening filter and contrast enhancement. Secondly, K-means clustering algorithm is used for preprocessing. Thirdly, feature extraction includes 2-dimensional discrete wavelet transform and features quantity reduction process includes principal component analysis. Fourthly, artificial neural network is used for classification of Glioma, Meningioma and Pituitary type tumors. An effective training function `Levenberg-Marquardt' is used for proposed network construction. This proposed method provides 95.4% accuracy, 94.58% sensitivity, 97.83% specificity. This improved result is comparatively better than other existing detection techniques. The basic reasons for getting magnificent results are the utilization of perfect preprocessing steps and effective training function.