Development of effective machine learning approaches for denoising medical images and their performance investigation

Abstract

Nowadays, physicians or radiologists require a best medical image quality for achieving efficient and fast diagnosis. Medical images generally have a problem of presence of noise during its acquisition and transmission reception, storage and retrieval. Noise corrupts the medical images and degrades the quality of the images. This degradation includes suppression of edges, structural details, blurring boundaries etc. Removing noise from the original image is still a challenging problem for researchers. Most of the researcher designed effective algorithms for noise removal. Process of image denoising or image restoration is still the most fundamental, largely unsolved and widely studied problem. In this first research work, an effectual Recurrent Neural Network (RNN) with long short-term memory based batch normalization is introduced for medical image denoising. Initially, the CT lung images with noises are taken as an input. The input image is denoised by using RNN. Batch normalization is a recently popularized method for accelerating the training of deep feed-forward neural networks. The Long Short-Term Memory (LSTM) involves batch normalization and demonstrates that doing so speeds up optimization and improves generalization. In batch normalization, the batch size is optimally selected by using Particle Swarm Optimization (PSO) algorithm. The proposed system was implemented using MATLAB. The experimental results are compared with the existing system. The Particle Swarm Optimization (PSO) algorithm has a long training time. To solve this problem the Firefly Algorithm (FA) is utilized for selecting an optimal batch size. In this second research work, the proposed newline