Revealing Brain Tumor Using Cross-Validated NGBoost Classifier

Abstract

Brain is the most complicated and delicate anatomical structure in human body. Statistics proves that, among various brain ailments, brain tumor is most fatal and in many cases they become carcinogenic. Brain tumor is characterized by abnormal and uncontrolled growth of brain cells, and takes up space within the cranial cavity and varies in shape, size, position and characteristics viz., can be benign or malignant, which makes the detection of brain tumor very critical and challenging. The vital information a neurologist or neurosurgeon needs to have is the precise size and location of tumor in the brain and whether it is causing any swelling or compression of the brain that may need urgent attention. This paper exploits ensemble strategy based Machine Learning (ML) algorithms for reveling brain tumors. NGBoost algorithm along with 5-fold stratified cross-validation scheme is proposed as classifier model that automatically detects patients with brain tumors. The proposed method is implemented with necessary fine-tuning of parameters which is compared against ensemble based baseline classifiers such as AdaBoost, Gradient Boost, Random Forest and Extra Trees Classifier. Experimental study implies that proposed method outperforms baseline models with significantly improved efficiency. The interfering features those have impact on brain tumor classification are ranked and this ranking is retrieved from the best classifier model.