Robust enhancement technique of 2D electrophoresis images
Nicolas Nafati
Michel Samson
Bernard Rossi
Plate-Forme Protéome-Pasteur.
Faculté de Médecine.
Nice. France
Abstract
6
Medical research aims to promote the life and health sciences related directly or indirectly to medicine. Discovery of new medicines is a determining element of this rsearch.
Now, the proteomic analysis (following after the genomics) became an important tool of the medical research involving top-level of modern technology measurement systems (densitometric imaging, mass spectrometry, etc. … ).
The Proteomic analysis always begins with separation of proteins. The most successful technique is: a two dimensional electrophoresis on polyacrylamid-gel. This technique consists in separating proteins according to both their charge and their mass. Protein spots appear under various profiles when gels are colored.
The main objective of proteomic analysis is to be able to track down proteins whose expression is modified in certain conditions, and then to identify them by mass spectrometry. The location of these protein candidates requires image acquisi-tion and then electrophoretic profile’s image processing. It is a heavy step in term of time, because existing software packages on the market do not allow a reliable detection of protein spots. This consequence is due to the noise, trails and artefacts generated by protein separation and the acquisition system.
In the 2D-gel electrophoresis field, the image quality is very importance. There-fore, the reduction of artifacts and noise, the improvement of the image quality is a key challenge.
In this paper, we describe an image processing technique which allows to improve the quality of the 2D electrophoresis images in term of contrast and enhancement, and consequently, to improve the reliability of protein detection. This technique consists in combining both Multi Scale Retinex Method (MSRM) and Subbands Decomposition/Reconst-ruction (SDR) technique.
Keywords
Proteomic analysis. Two dimensional electrophoresis on polyacrylamid-gel. Contrast. Enhancement. Multi Scale Retinex Method. Subbands Decom-position /Reconstruction technique.
1. Introduction
In the proteomic domain and before the peptide analysis, first, the biologists realize 2D gels by using the 2D electrophoresis technique. This one consists in migrating proteins according to the iso electric point (horizontal view), and then according to molecular mass (vertical view). The obtained gel is scanned by a densitometer system. The obtained images are always accompanied by different parasites: noise, artefact, trails, air bubble tracks [9].
In this paper, one will expose the principles of the decompostioin and reconstruction technique and that of multi scale retinex method. We give the effect improvement of retinex and subbands combination in term of protein spots detection.
2. What do we mean by the image quality ?
From an experimental point of view, the biologists often realize several gels under the same and/or different conditions, and from the same and/or different samples; The obtained gels often contain parasites such as air bubbles, cracks, trails, background, and artefacts. All these parasite signals lead to a bad and complex protein quantification [9] (Figure_1).
So, in proteomic image analysis context, the notified noise must be minimized in order to avoid parasite spots detection and Consequently to optimize the proteomic image analysis cost.
Generally, a good quality of image is translated by a good Signal to Noise power Ration (SNR), or by good Maximum Signal Power to Noise power (PSNR). In other words, less is the background noise, better is the contrast.
Figure_1: 2D-gel electrophorersis image. Various sorts of noises : artefact, trails, background.
3. Principle of the decomposition and reconstruction subband technique
In many image processing applications, the image is decomposed into a set of sub- images, and the information within each sub-
image is processed more or less independently of that in the other sub-image. This decomposition process is called the multi-resolution analysis and is a sort of image viewer at different scales and resolutions.
Figure_2.1 : Principle of subband decomposition. Rows decomposition of I(m,n).
Figure_2.2: Columns decomposition of IL(m,n).
Figure_2.3: Columns decomposition of IH(m,n).
As indicated in Figure_2, the input image is split into 4 sub-images : {ILL, ILH, IHL, IHH}, the first and last sub-image signal contains respectively the only low and high frequencies, ILH is the sub-image with horizontal low and vertical high frequencies, IHL is the sub-image with horizontal high and vertical low frequencies.
Figure_3 shows the result of this subband decomposition process.
Figure_3: One level subband decom-position. From top to bottom, clockwise, the obtained sub-images are {ILL1, ILH1, IHL1, IHH1}.
Also, if we decompose the obtained low frequencies sub-image ILL, into four sub-images, we obtain the following figure:
Figure_4: Two levels subband decom-position. The previous sub-image of low frequencies ILL1 is decomposed into these four sub-images {ILL2, ILH2, IHL2, IHH2}.
Generally, subband decomposition is realized by making an anti-aliasing numerical filtering followed by a down- sampling. In our case, the factor of this sampling is 2 (dyadic subband decomposition). The opposite of this decomposition process is the subband reconstruction which consists in
6
reconstructing the origin image from the four sub-images of the same level. Reconstruction operation is translated by dyadic up-sampling, interpolation of each sub-image, and finally addition of the obtained sub-images.
The filters that one used here, are wavelets 'db5' filters. Generally, wavelet filters allow a perfect decomposition and reconstruction [4][8].
4. Overview of the original Multi Scale Retinex method
The Retinex image processing was originally developed to improve the perception of images by the human eye. Elaborated on psycho-visual experiences, model Retinex has neither scientific justification, nor rigorous mathematical justification. His detractors are numerous, the efficiency on a lot of examples was noticed. We wanted to use it to improve the image enhancement [1,2,3], and then to intensify the protein spots.
Single Scale Retinex principle consists of illuminating an image by a white and constant light on all the image. Multi Scale Retinex is explained easily from SSR (SSR) [5,6,7]. For SSR we have:
where R(x,y,c) is the image output, I(x,y) is the image value, Ä denotes the convolution product, and F(x,y,c) is a Gaussian surround function explicitly
given by:
with K selected so that:
In the above, the constant "c" is the scale.
The MSR output is simply the weighted
sum of several SSR's with different scales.
The following figure (Figure_5) is the result of Retinex algorithm. The problem is that the image result tends towards a saturation grey level and therefore a loss of contrast. One observes nevertheless an interesting relief effect. One way suggest not to apply Retinex to the image on its whole but only to certain frequencies.
Figure_5: Result of applying Retinex method on the initial image ( Figure_1).
5. Subband and Retinex combination
For reasons of confidentiality of the project in which the algorithm was used, we can not regrettably give details of the subband and Retinex combination process. Nevertheless the basic idea is not to apply Retinex to the origin image, but only to certain frequencies selected by subband technique.
The results of this combination algorithm is given below in Figure_6. Figure_6.1 shows the origin image, Figure_6.2, the result of the combination process. One notices the contrast improvement and background minimization. Also, one remarks, the conservation of the image details and the protein spot enhancement.
Figure_6.1 : Origin image
Figure_6.2: Combination process. Reconstructed image from one levels decompostion subband
Figure_6.3: Combination process. Reconstructed image from two levels decompostion subband
6. Conclusion
The previous results show that the preprocessing image by retinex and subband combination algorithm leads to more effective detection of protein spots. This is essentially due to the improvement of the image quality in term of the SNR.
7. Références
[1] B. V. Funt, K. Barnard, M. Brockington, and V. Cardei, "Luminance based multi-scale Retinex," Proceedings AIC Colour 97, Kyoto, Japan, May 25-30 (1997).
[2] Brian Funt, Florian Ciurea, and JohnMcCann. Retinex in MaLab. Color Science. Systems and Applications. Pp 112-121. 2000
[3] Brian Funt. Kobus Barnard. Michael Brockington. Vlad Cardei. Luminance-Based Multi-Scale Retinex. Simon Fraser University. Proceedings AIC Colour 97 Kyoto 8th Congress of the International Colour Association, May 1997.
[4] Rafael C. Gonzalez and Richard E.Woods. Digital Image Processing (2 Edition). International Edition. 2002.
[5] D. J. Jobson, Z. Rahman, and G. A. Woodell, "Retinex Image Processing: Improved Fidelity To Direct Visual Observation". Proceedings of the IS&T/SID Fourth Colour Imaging Conference: Colour Science, Systems and Applications, Scottsdale, Arizona, November, pp. 124-126, 1996.
[6] D. J. Jobson, Z. Rahman, and G. A. Woodell, "Properties and Performance of a Center/Surround Retinex," IEEE Transactions on Image Processing, March 1997.
[7] D. J. Jobson, Z. Rahman, and G. A. Woodell, "A Multi-Scale Retinex For Bridging the Gap Between Colour Images and the Human Observation of Scenes," IEEE Transactions on Image Processing: Special Issue on Colour Processing, July 1997
[8] Nicolas Nafati. Contribution à l’étude du filtrage, de la décomposition en sous-bandes et de la reconstruction appliqués au traitement des images. Thèse de doctorat. CNAM de Paris 1995.
[9] Pierre Nugues. Interprétation de gels d’électrophorèses 2D. Thèse de Doctorat. Unversité de Nancy. 1989.
[10] Zia-ur Rahman, Daniel J. Jobson, and Glenn A. Woodell. A Multiscale Retinex for Colour Rendition and Dynamic Range Compression. SPIE International Symposium on Optical Science, Engineering and Instrumentation. Applications of Digital Image Processing XIX. Proceedings SPIE 2825, Andrew G. Tescher, ed., 1996.
[11] Z. Rahman, D. J. Jobson, and G. A. Woodell, "A Multiscale Retinex for Colour Rendition and Dynamic Range Compression", SPIE International Symposium on Optical Science, Engineering and Instrumentation, Applications of Digital Image Processing XIX, Proceedings SPIE 2825, Andrew G. Tescher, ed., 1996.
6