MR Brain 2D image Tumor and Cyst Classification Approach: an Empirical Analogy

Brain tumors and cysts might prove to be fatal if detection is delayed. Early detection of brain tumors and cysts ensures improved survival rates and thus has become a cause of concern. Multiple models using various algorithms have been used to make the detection accurate at an early stage. A cyst might be an alarming factor and might be the early stage of a brain tumor. This paper focuses on comparing the different methods used to detect brain tumors. The proposed model is trained and tested using six different algorithms. The Kaggle dataset has been used for both training and testing. Two classes from the dataset are used. One of the classes consists of Pituitary Tumors, while the other consists of No tumors. A total of 1222 images have been used for training the proposed model. Images have been resized into 200×200 dimensions for better results. Most of the tumors detected while testing is found to be pituitary tumors. We have used the ensembler classifier having six classes, Support Vector Machine (SVM), Logistics Regression, Random Forest, K Nearest Neighbors (KNN), Decision Tree, and Naive Bayes. Among all the algorithms, Logistic Regression shows the highest accuracy of 99.6%, followed by Random Forests and SVM. Thus, the paper helps us to determine a better algorithm among multiple algorithms for detecting brain tumors and to specify whether it is a pituitary tumor or not.