<![CDATA[The quality of Magnetic Resonance Imaging (MRI) images is often compromised by various types of noise, such as salt, pepper, salt-and-pepper, and speckle noise, caused by technical or environmental disturbances. This study aims to develop a brain MRI image denoising model based on the U-Net architecture, capable of effectively removing different types of noise. The methodology includes collecting normal brain MRI datasets, applying data augmentation to increase variability, and introducing artificial noise to simulate possible noise conditions. The U-Net model is trained and evaluated using the Mean Squared Error (MSE) and Peak Signal-to-Noise Ratio (PSNR) metrics. The novelty of this study lies in its combination of augmentation techniques, multi-intensity artificial noise variations, and its exclusive focus on normal brain MRI images. The results demonstrate that the U-Net model achieves optimal performance on salt-and-pepper noise at an intensity of 0.1, marked by the highest PSNR value of 37.2047 dB and the lowest MSE value of 0.000207. Conversely, the model shows the lowest performance on high-intensity speckle noise, indicating greater challenges in addressing multiplicative noise. This study contributes a systematic and empirically tested approach to improving the quality of brain MRI images with high efficiency, supporting the development of image-based diagnostic systems in the medical field.Keywords: Deep Learning, Denoising, Image Enhancement, Noise, U-Net.]]>
