A Comparative Study on Denoising of MRI Images and Classification of Brain Tumor Images Using Deep Learning Techniques

At present, Magnetic Resonance Imaging (MRI) is widely used for identifying and classifying brain tumor on humans. The American Cancer Society accounting 18,760 deaths in the past year occurred due to this deadly disease. To identify and diagnose tumor is crucial. The Deep learning (DL) technique, a subdivision of Artificial Intelligence (AI) contributes a vital role in medical image processing. It simplifies the analysis by avoiding computational complexity and time-saving. However, the analysis and classification depends on the images we used for training and testing. The images are recorded under environmental noise, motions of patient, and several other factors degrade the quality of image which may lead to losing of information on the image. To enhance the resolution of without losing data, a proper denoising technique is mandatory. For that, hybrid empirical mode decomposition (EMD) based total variation denoising (TVD) is proposed. It denoising the image and preserving sharp edges and data present in it. Furthermore, the U-Net based attention mechanism segment the tumor affected region and makes prediction more accurate. Using the segmented image, a comparative analysis of classification is done by utilising three distinctive neural networks. The performance evaluations of those models are discussed in word of recall, F1 score, accuracy, precision, PSNR, and SSIM. The proposed technique significantly improved the prediction rate and it outperformed the conventional techniques.