Salvaging tumor from T1-weighted CE-MR images using automatic segmentation techniques

When there exists a massive growth of abnormal cells inside the human brain, it is called as Brain Tumour. One of the major challenges in Content Based Image Retrieval (CBIR) is the semantic gap, which is extraction of the information between the human evaluator and Magnetic Resonance Imaging (MRI) machines. In the present scenario, the radiologist manually checks for the tumour region segmentation and outlines of that region. This problem can be reduced, if we combine the high-level and low-level feature extraction. In the proposed work, this gap is being bridged out, by using deep learning feature extraction technique over T1-weighted Contrast Enhanced-Magnetic Resonance (CE-MR) images along with Canny Edge, Gradient Descent Cost Minimization, Loss function evaluation and measurement of similarity through distribution vector and closed form metric learning as an automation segmentation technique to obtain the more precise accuracy. Also, the mean accuracy precision determined by the proposed model is 94.6% for the several query images sets. By this, manual efforts could be minimized in identifying the granular details of the tumours present in the several parts of the brain by employing MR images. © 2022, The Author(s), under exclusive licence to Bharati Vidyapeeth's Institute of Computer Applications and Management.