Improved Accuracy of Band Detection in GASepo System for Quantitative Analysis of Images in Epo Doping Control

Transcription

1 Improved Accuracy of Band Detecton n GASepo System for Quanttatve Analyss of Images n Epo Dopng Control S. Štolc *, I. Bajla ** E-mal: umerstol@savba.s * Insttute of Measurement Scence Slova Academy of Scences, Dúbravsá cesta 9, Bratslava, Slovaa ** BU-Hgh Performance Image Processng Smart System Dvson Austran Research Centers GmbH - ARC, A-2444 Sebersdorf, Austra Abstract: In the year 2003, World Ant-Dopng Agency (WADA) ntated an nternatonal project wth the goal to develop a sutable and easy to use software system amed at the standardzed quanttatve analyss of dgtal mages acqured durng the process of erythropoetn (Epo) dopng control. Wthn the cooperaton between Austran Research Centers GmbH and Insttute of Measurement Scence, Slova Academy of Scences, mproved mage preprocessng method has been developed. The proposed method Band Straghtenng Algorthm (BSA) sgnfcantly reduces local geometrc dstortons n Epo mages that cause band detecton problems, especally ncases wth geometrcally dstorted and dsrupted bands. The accuracy of the band detecton s one of the most crtcal ponts of the Epo mage analyss. In the year 2006, the proposed method has been mplemented n the latest release of the software pacage GASepo 2.1, whch s used n Dopng-Control laboratores worldwde. Keywords: Epo dopng control, mage segmentaton, band detecton, mage flterng, geometrcal dstorton correcton, correlaton analyss 1. Introducton In the last decade, the synthess of recombnant forms of human peptde hormone erythropoetn (Epo) became possble due to advances of molecular engneerng. Ths substance was orgnally ntended for therapy of dseases assocated wth lac of endogenous human Epo whch controls the producton of red blood cells. In a short tme the recombnant erythropoetn (repo) turned to be a popular dopng substance especally n endurance sports where performance boost of up to 10% can be acheved [1,2]. Unle the conventonal dopng substances (e.g. anabolc sterods) repo cannot be detected by conventonal mass-spectroscopy methods [4]. The research has shown that the dfference between repo and natve human Epo s hardly evdent as late as n posttranslatonal phases n urne of a doped ndvdual. Detecton of such small dfferences ss done by double-blottng method based on soelectrc focusng n the gradent of ph n polyacrylamde gel [4]. Afterwards, Epo soforms are made vsble by a chemolumnescence reacton and recorded by a standard analog or dgtal photographc technque (see Fg. 1). In Austran Research Centers, the unque software pacage GASepo (Gel Analyss System for Erythropoetn) has been developed and put nto the praxs n many WADA accredted dopng control laboratores. 14

2 Fg. 1. The example of an Epo mage where the global and local geometrc dstortons are present. Leftmost and rghtmost lanes represent samples of the dopng standards EPOα,β and DARBα. All other lanes represent samples taen from tested ndvduals. Fg. 2. The example of the manual correcton of the global (left) and local (rght) mage dstortons n GASepo. 2. Manual mage rectfcaton n GASepo Geometrc mage dstortons are the most often reasons of unwanted output errors. Therefore, t s essental to rectfy the most strng geometrc dstortons present n Epo mages before any further mage analyses. The global mage dstortons have to be corrected manually requrng nteractve nterventon of the user (see Fg. 2 left mage). Manual correctons of the local mage dstortons are usually too tme demandng and therefore unable to use (see Fg. 2 rght mage). 3. Epo dopng postvty crteron The man tas of a dopng postvty decson s to fnd a rght placements of Fg. 3. The result of the ntal part of an Epo mage analyss. The two reference cut-off lnes are placed accordng to the dopng standards (lanes 1 and 9). two reference cut-off lnes that separate the groups of EPOα,β and DARBα bands from the remanng bands reflectng natural Epo (see Fg. 3). The fnal dopng postvty crteron counts wth exact postons of all bands that are compared wth the postons of cut-off lnes. Fg. 4. The basc scheme of the band detecton process. Schematc embeddng of the new Band Straghtenng Algorthm (BSA) s mared wth the blue color. 15

3 4. Epo mage segmentaton and classfcaton Due to the mentoned Epo dopng postvty crteron, the crucal pont of Epo mage processng s the procedure of the band detecton. Ths procedure conssts of two man steps mage segmentaton and object classfcaton. The segmentaton s a multstage process n whch the orgnal grayscale mage s transformed nto a bnary map [5]. Detected objects are joned together n the projecton operaton (PRJ) and subsequently classfed nto two groups band and artfacts [6] (see CLASS n Fg. 4). These operatons are hghly senstve to qualty of nput mages and therefore they ntroduce ultmate sources of output errors (see Fg. 5). 5. Band Straghtenng Algorthm (BSA) For suppressng unwanted local geometrc dstortons present n Epo mages, a new mage preprocessng method has been developed. The man dea of Band Straghtenng Algorthm (BSA) s automatc search for an optmum vertcal shfts of all mage columns that lead to maxmum reducton of the local geometrc dstortons (see Fg. 6). The search s performed n two successve phases the lnear search followed by the gradent descent search. Both methods mnmze the total row-wse tenson of an nput mage. An optmum shft vector opt s calculated accordng to the followng formulaes: opt M ~ R ~ = arg max{ M ( L )}, m ~ ~ ( L ) = Tens( R ), Tens n ~ ( R ) ( ~ ~ = a, j a, ( j 1) ) = [ a~, a~, K, a~ ],,1 = 1 j= 2,2, n 2, Fg. 5. The example of an ncorrectly segmented Epo mage n consequence of strong local geometrc dstortons. One can see the objects consstng of wrong object fragments joned by the confused projecton operaton. The joned objects are shown n the same color ndex. The desred correct objects are bordered by the sold lne. Fg. 6. Basc prncple of BSA schematcally explaned on a model Epo mage. Local geometrc dstortons are beng reduced by the automatc search for an optmum vertcal shfts of all mage columns. Fg. 7. The example of an Epo mage processed wthout (a, b) and wth (c, d) the use of BSA. One can see the clear straghtenng effect of the source mage usng BSA (c vs. a) as well as the remarable mprovement of correct mergng of related object fragments (d vs. b). 16

4 where L ~ s the shfted Epo mage by a shft vector, M ( L ~ ) s the maxmzed cost functon, Tens ( R ~ ) s the row tenson of the -th row R ~. The detaled descrpton of the algorthm s gven n [8, 9]. Ths approach provdes substantal mprovement of mage qualty on acceptable computatonal complexty level. The experments wth real Epo data have shown that BSA s capable of lowerng average classfcaton error up to 2-4 tmes n computatonal tme O(mn 3 ), where m and n are heght and wdth of processed mage (see Fg. 7). 6. BSA mplementaton n GASepo In 2006, the novel BSA method was successfully mplemented n the latest release of the software pacage GASepo 2.1 (Fg. 8). Ths mplementaton requred: frst, to prepare a prototype of BSA and ts careful testng n Matlab envronment, then ts optmzaton n Mcrosoft Vsual C# 2005 tang advantages of Mcrosoft.NET Framewor Conclusons Snce a number of Epo mages suffer from geometrc dstortons whch affect relevant operatons appled to them n GASepo software, a new mage preprocessng method Band Straghtenng Algorthm has been developed. Ths method enables substantal reducton of the local geometrc dstortons resultng n the decrease of average classfcaton error by up to 2-4 tmes whereby the computatonal complexty stays on acceptable level maxmum tme O(mn 3 ). In 2006, the proposed method has been mplemented n the latest release of the software pacage GASepo 2.1, whch s used n DC labs worldwde. Fg. 8. Screenshot of the new GASepo Acnowledgements The wor on BSA algorthm development reported n the paper has been carred out wth the support of World Ant-Dopng Agency, Slova Grant Agency for Scence (Grant No.: 2/4026/04) and FP6 Project BAMOD (Contract No.: ). 9. References [1] Bajla I. et al.: Analyss of EPO Images after Isoelectrc Focusng and Double Blottng. In: Proceedngs of the Second Intern. Conference on Bomedcal Engneerng IASTED 2004, February 2004, Innsbruc, Austra, p [2] Bajla I. et al.: GASepo a software soluton for quanttatve analyss of dgtal mages n Epo dopng control. Computer Methods and Programs n Bomedcne 80(3), 2005, p [3] Holländer I. et al.: GASepo System for Analyss of Images Generated n Epo Dopng-Control. In: Proceedngs of the 17th Bennal Intern. EURASIP Conference Bosgnal 2004, June 2004, Brno, Czech Republc, VUTIUM Press, p [4] Lasne F., de Ceaurrz J.: Recombnant erythropoetn n urne. Nature, 2000, p [5] Ramoser H. et al.: Segmentaton of Electro-phoretc Images n Dopng Control. In: Proceedngs of the Intern. 17

5 Conference on Mathematcs and Engneerng Technques n Medcne and Bologcal Scences METMBS 2004, June 2004, Las Vegas, USA, CSREA Press, p [6] Wtovsý V. et al.: Alternatve Approaches to Band Classfcaton n Epo mages for the GASepo software. Techncal report ARC IT 0115, January 2005, ARC Sebersdorf research GmbH, 60 pp. [7] Štolc S., Bajla I.: Improvement of band segmentaton n Epo mages va column shft transformaton wth cost functons. Medcal and Bologcal Engneerng and Computng 44 (4), 2006, p [8] Štolc S., Bajla I.: Improvement of band segmentaton n Epo mages va column shft transformaton wth cost functons (Erratum). Medcal and Bologcal Engneerng and Computng 44 (7), 2006, p