International Journal of Recent Technology and Engineering (IJRTE) ISSN: 2277-3878, Volume-8 Issue-3, September 2019 2691 Published By: Blue Eyes Intelligence Engineering & Sciences Publication Retrieval Number: C4950098319/2019©BEIESP DOI:10.35940/ijrte.C4950.098319 Abstract: Microarray technology allows the simultaneous profiling of thousands of genes. Denoising is an important pre-processing step in microarray image analysis for accurate gene expression profiling. In this paper, as FFDNet provides model independent denoising technique, it is been applied on microarray images. FFDNet is validated on AWGN based images and real noisy images trained network. The application is compared with the standard denoising methods. The results revealed that optimal sigma value to efficiently remove noise while preserving details for AWGN based images and real noisy trained methods were 15 and 20 respectively. Overall, the performance of the FFDNet is better compared to other metrics considered in the study as it is flexible, effective and fast. It is also capable to maintain the trade-off between denoising and feature preservation. Keywords : Microaray, Denoising, FFDNET, AWGN, Sigma Value, Performance metrics.. I. INTRODUCTION The Microarray technology can be visualized as a tool that opens the cell content, extracting its genetic content, identifying the genes that are activated and capable of generating those genes for purpose of analysis and reporting [1]. Microarray provides the multiple-gene expression simultaneously. Microarray finds its applications in the field of antibiotic treatment, in early cancer detection, early detection of oral lesions [2]. Typical microarray procedure for preparing a microarray slide involves sample preparation, labeling, hybridization, washing, image acquisition, and data analysis. After image acquisition, they are preprocessed before the data analysis. The preprocessing pipeline of microarray images typically includes gridding and spot-fixing, segmentation and intensity extraction [3]. Since the advent of microarray, noise has been an inherent part of microarray images. Typically, major noise in the microarray images is due to the optical procedures involved in the preparation of the sample slides and the complex biochemical reaction-taking place during acquisition [4]. The noise in the images makes it tedious for decision making during the data analysis. As based on the data analysis most of the decisions for respective applications are made. This makes denoising of the microarray images as one of the critical parts of the preprocessing pipeline. Two well established models in microarray denoising are transform domain approach and Revised Manuscript Received on September 15, 2019 Priya Nandihal*,ISE department, SDMCET,Dharwad,India. Email: talk2priya.nandihal@gmail.com Vandana S Bhat, ISE department, SDMCET,Dharwad,India. Email: vsreenivas6@gmail.com Jagadeesh Pujari , ISE department, SDMCET,Dharwad,India., Email: jaggudp@gmail.com spatial filtering. Recent trends have shown significant development in the field of machine learning/deep learning [5] based methods that are mainly of two kinds, supervised learning methods and unsupervised learning methods. The main advantage of these methods lies in the learning feature that they have incorporated. Based on the extent of training using the existing data determine the performance of the supervised learning algorithm and the extent of learning based on the data acquired for a particular application determine the efficiency of the unsupervised learning algorithm. Numerous researchers have contributed to the field of denoising of microarray images. Some of the noted contributions are mentioned to provide a wide range of methods developed in this field. In order to remove the random noise in the images, a stationary wavelet-based method was developed[6]. Utilizing optimize spatial resolution (OSR) and spatial domain filtering (SDF) denoising of microarray images was performed[7]. A framework was developed in order to remove both additive and multiplicative noise in the images as proposed in [8]. Fuzzy filtering based method for denoising of microarray-based images was developed [9]. Utilizing a peer group concept a switching scheme based on the impulse detection mechanism was developed [10]. Coefficients of subbands in the wavelet domain were employed for smoothening the noise levels was developed in [11]. FFDNet is a recent technique based on the neural network based method[12]. The main advantages of FFDnet are it is a fast and flexible denoising neural network. FFDNet is capable of dealing with the noise spatially variant noise and noise of different levels. In order to maintain the trade-off between the noise reduction and detail prevention noise level maps are provided. FFDNet exhibit potentially appealing results on both synthetic noise and real-world noise. The features exhibited are not present in various methods mentioned in the above paragraph. II. MATERIALS AND METHOD FFDNet provides various variants of noise models. Two types of models explored for microarray images utilizing the FFDNet: Additive White Gaussian Noise (AWGN) based model and real images based model. The network employed for denoising of the microarray-based images is shown in figure 1. The noisy image is downsampled into four sub-images. The sub-images are concatenated with the noise level map M to form a tensor of size Application of FFDNET for Image Denoising On Microarray Images Priya Nandihal, Vandana Sreenivas, Jagadeesh Pujari