Enhancing Brain MR Image Quality Using CNN With Best Denoising Modality for Improved Diagnosis of Abnormality: An Appraisal

Abstract

Digital medical images acquired from the brain are highly susceptible to noise, which causes significant challenges for radiologists to identify abnormalities in a precise manner. Noise interference hampers both diagnostic accuracy and the interpretation of underlying abnormalities, potentially leading to flaw conclusions. Magnetic Resonance (MR) imaging is the most preferred digital imaging technique for brain abnormality detection. To achieve precise detection, noise-free MR images are essential. Denoising modalities commonly address this issue by reducing unwanted noise while preserving essential image features. However, the effectiveness of denoising methods varies, and achieving an optimal filtered denoised image remains a challenge. This paper undertakes a thorough appraisal of various prominent denoising techniques on two public MR image datasets. The result shows Anisotropic Diffusion Unsharp Masking Filter (ADUM) as the most effective denoising method. A hybrid method that combines a Convolutional Neural Network (CNN) with ADUM filters is proposed to enhance feature extraction and abnormality detection of brain MR images. The performance of these methods is comprehensively evaluated through both qualitative and quantitative measures. The result shows that the proposed method does a better job of reducing noise while keeping edges than other conventional denoising methods, as shown by the examination of the results. This makes it a promising tool for both clinical and research use.