Segmentation Of Two-dimensional Gel Electrophoresis Image Based On Markov Random Field

Two-dimensional gel electrophoresis technology as a core protein analytical tool plays anextremely important role in the development of proteomics and it also grows quickly alongwith the progress of proteomics. The two-dimensional gel electrophoresis image usuallyregards gel as carrier and separates the proteins as spots on this gel according to the isoelectricpoint and molecular weight of the proteins. The analysis system which consists of three parts:image acquisition, image processing and the assessment of different protein in a gel image.On account of the large amount of data, gel image analysis (including imagepreprocessing, protein detection and segmentation, points matching and so on) must be donewith the help of computer analysis. So the computer-based analysis of gel images becomes acrucial step in proteomic analysis.In generally, it can make the extraction of target feature and parameter measurementpossible when the image segmentation is successful. Besides it can also promote thedevelopment of image analysis and image understanding of higher level. Effectivesegmentation methods can reduce the over-segmentation and under-segmentation of an image.At the same time, they can pick up the required features areas extremely. In addition to this,they distinguish the protein cross point effectively, helping improve the reliability of pointmatching results and protein which was determined. At last, they also lead to more accurateestimation of spots properties, e.g. spot grayscale and volume that realize the biomarkers ofgel image. Markov random field is based on Markov random field model and theBayesian theory, which provides the link between the prior knowledge and the uncertainty ofdescription. And according to the characteristics of the observed image, it determinesthe objective function of segmentation problem by using estimation theory and statisticaldecision in some optimal criterion. Finally, it can determine the maximum possibledistribution to satisfy these conditions, which will be divided into optimization problems. Thisarticle which is mainly on the basis of Markov random field, aims at exploiting the imagesegmentation for gel images.In the traditional segmentation algorithms， the concept that based on thresholdsegmentation algorithm is relatively simple，and the computation efficiency is higher. For the image of stronger target and background，the algorithm based on thresholdsegmentation has a good processing method， but it depends more on gray level. Thealgorithm based on region segmentation can take into account the spatial relationship of theimage，but it emphasizes the region’s overall properties，which easily lead to undesirableimage segmentation boundaries. And because it belongs to a kind of iterative algorithm, thespace and time it costs is relatively large. Though the algorithm based on edge detection has agood segmentation result for the edge positioning accuracy，detection precision and thedetermination of boundary. But it is sensitive to noise，and there’s not a standard measurementfor edge detection results of different algorithms. Besides that, watershed algorithm is alsoused for the gel image segmentation in advantage of its fast calculation and accurate edgeorientation. But it must be paid attention to the image over-segmentation. The method basedon the Markov random field can get the better result, because it not only considers the pixelspace correlation fully, but also can be able to describe each pixel that belongs to categoryclassification and the surrounding pixels between interdependent relationships.This article proposes a segmentation algorithm which is based on an improved Markovrandom field. In order to realize the preprocessing of the images, it uses the Non-Local meansalgorithm firstly. Then, with the help of Bayesian theory, it gets the results of the imagesegmentation. It must be noted that this process obtains the prior probability and thereforegets a posteriori probability by a second-order logical model (MLL). Meanwhile, italso introduces the gray density weight to update the clustering and variance, finally realizesthe optimization of gel image segmentation. Simulation results show that the improvedalgorithm is superior to original algorithm, it has overcome the detection and segmentation ofthe weak point and cross point to some extent. So it will further improve the accuracy of thegel image segmentation.