Optimal Denoising System for Medical Images Using Recurrent Neural Network and SVM

Abstract

Image denoising serves as a crucial preprocessing step in the realm of medical image analysis, with the primary objective of faithfully reconstructing the original image from its noisy counterpart. This process is essential for maintaining the integrity of vital details, such as edges and textures, within the denoised image. Innovatively addressing this challenge, our proposed system introduces a novel approach that seamlessly integrates Recurrent Neural Network (RNN) and Support Vector Machine (SVM). This powerful combination is adept at efficiently eliminating various types of noise, including gaussian, white noise, salt and pepper noise, and speckle noise, from intricate lung CT images. To enhance both learning accuracy and training efficiency, we have incorporated batch normalization in conjunction with residual learning. Notably, batch normalization is executed with the support of Long Short-Term Memory (LSTM). This strategic integration aids in the gradual separation of image structure from the noisy observations, a pivotal aspect in achieving optimal denoising outcomes. This approach not only enhances the accuracy of denoising but also contributes to reducing the overall training time, making it a valuable advancement in medical image preprocessing.