Automatic Classification of Brain MRI slices into Glioma/Meningioma using Deep Transfer-Learning
A. S. Vickram, Mathan Muthu C M
Abstract
Brain tumor (BT) is a harsh brain abnormality and early detection and treatment is essential to reduce the impact of the disease. This work proposed a Deep Transfer-Learning (DT) approach to detect the Glioma/ Meningioma with a better accuracy. The stages of this work includes; image collection and modifying its dimension to 224x224 pixels, feature extraction using chosen DT-models, feature reduction and serial features integration to get the fused-features, and bi-level classification using SoftMax and verifying the merit using 5-fold cross validation. This work initially executed the detection using individual features and then the experiment is repeated using the fused-feature vector (FFV). The FFV based detection helps to achieve a detection accuracy >97% on the chosen task and this outcome confirms the clinical significance of the proposed technique. In this work, the detection is performed using only the axial-plane 2D MRI slides, and in future, it can be extended to clinical MRI examination task.