Use of Discrete Cosine-based Stockwell Transform in the Binary Classification of Magnetic Resonance Images of Brain Tumor

Abstract

biomedical diagnostic tool for the detection of tumors in the brain since it provides detailed and comprehensive information associated with the brain's anatomical structures. The radiologist can detect the existence of malignancies or aberrant cell growths using MRI images. Early-stage brain tumor diagnosis and treatment are greatly aided by MRI image processing. This study inquires about a method for classifying MRI brain images into without tumors and brain tumors to detect tumors using these images. These days, researchers can create reliable Computer-Aided Diagnosis (CAD) systems for identifying tumors and healthy brains thanks to the benefits of machine learning. A crucial stage in any machine-learning model is feature extraction. Time-frequency analysis techniques are more effective for image classification applications since they provide localized information. We suggested using the Discrete Cosine-based Stockwell Transform (DCST) to extract the efficacious features from brain MRI images and create the feature matrix after pre-processing and segmentation. The feature matrix's dimension was decreased using the chi-square test. A Support Vector Machine (SVM) classifies the selected features at the end. We employed a dataset containing 7023 brain MRI images divided into four classes: tumors of the pituitary, glioma, meningioma, and without tumors. For binary classification into brain tumors and no tumors, we attained an accuracy of 97.71%.