Research On The Detection Technology Of Consensus Protein Spot Set In 2DGE Images

Proteomics is the large-scale study of proteins from a tissue or cell,to explore the relationship between the dynamic changes of the proteins and the pathological or physiological process of living organisms.Proteomics could provide a basis for the research fields such as clinical diagnosis,treatment,drug development,environment pollution analysis and so on.Two-dimensional gel electrophoresis(2DGE)is one of the most valuable methods for the proteome separation.Consensus protein spot set(CPSS)detection is the core technology for 2DGE.In this thesis,under the support of the project of National Natural Science Foundation of China,the key technologies of CPSS detection such as elastic image registration,preprocessing,protein spot detection and matching are intensively studied.The specific research work and the achievements obtained include:(1)To improve the CPSS detection accuracy,an innovative elastic 2DGE image registration method based on CSIFT feature is proposed.Firstly,SIFT(scale-invariant feature transform)points in each 2DGE image is detected.Then,the SIFT points are classified into inliers and outliers by projection transform model,and the outliers are divided into two subclasses as true outliers and false outliers.At last,inliers and false outliers are used as the control points of the TPS transform to implement elastic registration.The data redundancy of intra-group 2DGE images is used to improve the CPSS detection.Experiments on different 2DGE images show that this method can implement robust registration regardless of image scale and rotation,and is robust across a substantial range of affine distortion,change in 3D viewpoint,addition of noise and change in illumination.Compared with the traditional image registration methods with protein spot landmarks,the proposed method also can improve the registration precision up to94% and enhance the automation without manual parameter inputs.(2)To overcome the false detection of protein spots caused by impulse noise and artificial interference,such as lanes,cracks,hair,fingerprint and dust,a novel approach based on local minimum classification is proposed.Firstly,it detects the local minima with h-basin transform.Secondly,it discriminates impulse noise and artificial interference from protein spots according to the neighbouring grad difference.Then,it restores the noisy pixels by directional average filtering.At last,it iterates the detection and filtering process until there's no noisy minimum.Experiments on synthetic and real 2DGE images show that,compared with the traditional preprocess methods,the propose method can greatly reduce the damage to the useful information of the image.From the three indicators of m SSIM objective evaluation,protein spots center minimum correct detection rate and subjective evaluation,the proposed method is comprehensively superior to the existing methods.The correct detection rate of protein spot centers in unfiltered 2DGE images is only66.31%,and it rises up to 98.12% in the images filtered with the proposed method.To reduce the quantization error caused by illumination difference in 2DGE images,a novel image gray correction algorithm-neighboring grad section histogram equalization(NGSHE)is proposed.NGSHE can solve the over-enhance problem in background regions.To reduce the quantization error of protein spots caused by uneven background,a background subtraction method based on surface fitting is proposed.Since Gaussian noise would lead to the spot edge location deviation and quantization error,the adaptability of various Gaussian noise filters for 2DGE image is contrastively studied.We can conclude that Gaussian filter is the most applicable filter for 2DGE images.(3)Aiming at the inaccurate segmentation of protein spots caused by inaccurate internal and external markers of the traditional marker controlled watershed transform(MCWST),an improved MCWST method is proposed.Firstly,hierarchical h-basin minima are extracted as the internal markers,hence increase the detection accuracy of protein spot centers.Secondly,an optimized external marker extraction method based on the fusion of grayscale and distance markers is put forward.The experimental results show that the correct detection rate of protein spots in this method is more than 89.10%,which is obviously higher than that of the traditional MCWST method.The improved method works well for individual protein spots,but it still fails to separate heavily overlapping protein spots.Hence,the valley characteristic based method is proposed to separate overlapping spots.The separation lines of overlapping protein spots are determined by extracting skeleton,establishing topological structure and scanning valley characteristics.The experimental results show that this method can effectively separate the severely overlapping protein spots,which can not be separated by traditional separation methods.(4)To improve the accuracy of protein spot matching,a progressive and cooperative protein spot matching method is proposed.Firstly,after similarity graph analysis,the spot center coordinate,the spot gray value and the shape context feature are selected as matching features.Secondly,the confirmed protein spots are divided into several grayscale bands,and after calculating out the distances of the coordinates in the same grayscale band in the registered image,k-nearest spots are selected as candidate matching spots.Thirdly,the final matching spot are determined according to the maximum shape context similarity between the target spot and the candidate match spots.The matched spot pairs are to be used as landmarks,while the unmatched spots were pushed into the queue of suspected spots.Finally,the suspected spots are collaboratively verified based on multi-image,multi-spot and multi-feature.The experimental results show that the proposed method can effectively overcome the undetection of spots caused by the premature use of the thresholds in traditional methods,and the correct matching rate of protein spots is about 98%.The above achievements have been applied to the first domestic gel image analysis software Protein Master1.1 which is developed by our research group.The software has been applied in some universities and companies and achieved valuable results.The research results of this thesis provide a new set of theory and methods for 2DGE image analysis.