Performance Evaluation of Medical Image Denoising based on Deep Neural Network

Medical image denoising plays a critical role in enhancing the quality and interpretability of diagnostic images, impacting the accuracy of clinical diagnoses. This study focuses on the performance evaluation of medical image denoising based on Deep Neural Networks (DNNs). The investigation emphasizes the pivotal phases of denoising, including data preparation, network architecture design, training, and optimization. The denoising technique is conducted, considering the challenges posed by various types of noise (Gaussian and Poisson) commonly present in medical images. The proposed DNN consists of several layers where, the input layer contains noisy medical images which are fed to 2D convolution layer as an input. Spatial down sampling is achieved using 2D MaxPooling, which lowers the dimensionality of the input data while retaining the most crucial features. After achieving pooled features, the up sampling has been employed which results the denoised image. The loss function has been calculated by measuring mean square error (MSE). Furthermore, the performance measures for proposed DNN have been evaluated in terms of PSNR and SSIM. When compared to Gaussian noise, the proposed method has experienced an improved performance of 4.5% in terms of PSNR AND 29.34% in terms of MSE.