Features Driven Brain Tumor Detection Using Machine Learning Models

Brain tumor detection seeks to separate tumorous brain images and healthy images, which is considered to be the first step in diagnosis in order to maximize the likelihood of successful tumorous region detection accuracy. Magnetic Resonance Imaging (MRI) is a viable input in brain tumor anatomy, which helps it in making an effective diagnosis system. To handle this research issue, a feature driven brain tumor detection model is proposed which takes BraTS 2015 MRI images dataset as input and provides segmented result label as output. The dataset contains T1, T2, T1Gd, and FLAIR modalities for each patient. The input data is preprocessed, wherein it is first converted from the .nii.gz format to NumPy arrays, and then divided into 4 subsets namely HGG-1, HGG-2, HGG-3, and LGG-1. The images are then cropped to discard the unnecessary background and to narrow the object region for all the four modalities. Feature extraction and Grey Level Co-occurrence Matrix (GLCM) generation is then done based on the cropped image. Fourteen Haralick textual features are computed from the GLCM Matrix and standardization of the independent features using feature scaling presents data in a fixed range. Finally, supervised machine learning classification algorithms, namely Support Vector Machine, K-Nearest Neighbors, and Random Forest are used to train the model. Out of all, Random Forest algorithm reached the highest performance accuracy.