Machine Learning based Brain Tumor Detection and Classification using HOG Feature Descriptor
Swati Shilaskar, Tejas Mahajan, Shripad Bhatlawande, Suraj Chaudhari, Rohan Mahajan, Khushi Junnare
Abstract
The timely identification of brain tumor is crucial in medical diagnostics as it aids in achieving accurate diagnoses. Medical imaging techniques have witnessed significant advancements with the progress of engineering, and medical diagnosis can rely on Magnetic Resonance Imaging (MRI) as a reliable option. However, manually scrutinizing MRI images is a daunting task since they contain vast amounts of data and intricate details that are challenging for humans to recognize. Consequently, it is imperative to automate these techniques for efficient and precise analysis. The prior studies have mostly been concerned with identifying tumors using ML models trained on small datasets, which can sometimes produce relatively weak models that are less reliable. This study presents a proposed method to ease the process of brain tumor detection and classification with the use of MRI technology. Variations in the data make brain tumor detection a complex and challenging task. Preprocessing, Feature Extraction, and Classification are the three main components that constitute the proposed system. Preprocessing is used to eliminate noise from the raw data, while HOG feature descriptor is utilized for feature extraction. Various ML classifiers, including, Support Vector Machine (SVM), Gradient Boost, K Nearest Neighbor (KNN), XG Boost, and Logistic Regression are utilized to evaluate the system’s precision. The results demonstrate that different classifiers offered different levels of accuracy, with the Extreme Gradient Boosting (XG Boost) classifier providing the highest accuracy of 92.02%.