Brain Tumor Classification using Improved Gray Level Co-occurrence Matrix and CNN

Abstract

Automated tumor detection in medical imaging has been one of the developing fields in medical diagnostic applications. Anomalies of the human body are captured using different imaging techniques. The captured images must be understood and processed for further diagnosis and treatment planning of these anomalies. Although there are skilled medical professionals who will be able to understand the medical images and detect the anomalies. The limited availability of human experts and a large amount of data will cause a problem. In addition to that, there are chances that the diagnosis is prone to human error. This decreases the effectiveness of the diagnosis. Since Convolutional Neural Networks (CNN) is one of the best methods for image analysis, it is proved to be more effective. Therefore, CNN can be used for classification purposes. Automated tumor detection through MRI has become very informative as it provides information about abnormal tissues. This information becomes a necessity for planning treatment. To provide this information we have come up with CNN architecture for brain tumor classification consisting of three tumor types as well as no tumor. The analysis of brain Magnetic Resonance Imaging (MRI) is done mainly focusing on feature extraction and classification. A model is created using improved Gray Level Cooccurrence Matrix (GLCM) for Feature Extraction and CNN for Classification, which will help to classify the tumors with almost 100% accuracy.