Abstract
The death rate of humans due to the brain tumor was high some year before but due to the early diagnosis of brain tumor, this rate is significantly decreased. Due to the accurate brain tumor diagnosis on early stages, long survival chances for a patient is increased. In this paper, we proposed computationally efficient and accurate brain tumor segmentation method. The proposed method is divided into different phases. In the first phase, the image is decomposing into wavelet sub-bands and then the high energy sub-band is divided into blocks. Then, high variance features from each block are selected through discrete cosine transform and passed to neural network for classification. The classification accuracy rate of 99.7% is achieved which is good as compare to existing ones.