an evaluation of airport x-ray backscatter units based on image characteristics

7/31/2019 An Evaluation of Airport X-ray Backscatter Units Based on Image Characteristics

1/33

An evaluation of airport x-ray backscatter units basedon imagecharacteristics

Leon Kaufman&Joseph W. CarlsonReceived: 27 October 2010 /Accepted: 9 November 2010 #Springer Science+Business Media, LLC 2010AbstractLittle information exists on the performance of x-ray backscatter machinesnow be ingdep loyed through UK, US and o ther a i rpor ts . We implement aM on te C a r l o s i m u l a t i o n u s i n g a s i n p u t w h a t i s k n o w n a b o u t t h ex - ray spec t ra used f o r imag ing , dev i ce spec i f i ca t i ons and

avai lable image s to est im at e penet rat i on and exposure to the body fromthe x-ray beam, and sensitivity to dangerous contrabandmaterials. We showthat the body is exposed throughout to the incident x-rays, andt h a ta l t h o u g h i m a g e s c a n b e m a d e a t t h e e x p o s u r e l e v e l sc l a i m e d ( u n d e r 1 0 0 nanoGrey per v iew), detection of contraband canbe foiled in these systems. Becausefront and back views are obtained,l ow Z ma te ri a l s can on l y be rel i abl e detec ted i f t h ey a r e pa ck edo u t s i d e t h e s i d e s o f t h e b o d y o r w i t h h a r d e d g e s , w h i l e h i g hZ materials are well seen when placed in front or back of the body, but not tothe sides.Even if exposure were to be increased significantly, normalanatomy would make a dangerous amount of plastic explosive with tapered

edges difficult if not impossibleto detect. KeywordsX-ray imaging .Airportsecurity.Comptonbackscatter .Explosivedetection.PersonnelscreeningIntroductionA i r p o r t w h o l e b o d y s c a n n e r s a r e b e i n g d e p l o y e d t h r o u g h o u tt h e US , UK , so me a i rpo r t s i n con t i nen tal E u rope and o the r pl aces i nthe wor ld . The purpose o f these

J Transp Secur DOI 10.1007/s12198-010-0059-7L. KaufmanUniversity ofCalifornia, San Francisco, Emeritus, 161 4th Ave, San Francisco, CA 94118,USAe-mail: [email protected] J. W. Carlson (*)Formerly University of California, San Francisco, 240 Cambridge Drive,Kensington, CA 94708, USAe-mail: [email protected]


2/33

units is to find contraband hidden under clothing but on the surface of thetraveler.One of the technologies in use involve x-ray scanning withbackscatter imaging.Referring to Figure1, in conventional radiography anobjec t o f interes t (M) whichma y i nc lu de an ad ded ab sor be r (A ) isp lac ed be t ween a sour ce o f x - r ays S a nd an imag ing detec to r I . The

detector produces a negative of the absorbing power of theM-Acombination. X-rays can scatter in M and A and reach the detector,degradingimage quality. A grid can be placed between M and I to reducescatter. In backscatter imaging, a source/collimator (S/C) assembly isscanned over the object of interest,and a non-imaging detector (N-I) isplaced between the col l imator and the object. X-ray s th a t un de rg oC o m p t o n s c a t t e r i n t h e o b j e c t m a y r e a c h t h e d e t e c t o r .Positioninformation is obtained by knowing where the x-ray beam is.Photons can undergo


3/33

F i g u r e 1

Schemati c compar ingconventional and backscatter x-ray imaging


4/33

scatter anywhere along the beam as they penetrate, and many may scattera secondt i m e a n d r e a c h t h e d e t e c t o r , b u t i n d e p e n d e n t l y o ft h e i r o r i g i n i n t h e o b j e c t , a r e a ss i gn ed t o t he b ea m p os i t i on .I f t h e b e a m s c a n s i n a p a t t e r n p a r a l l e l t o i t s e l f , t h e n t h e x , y ( i n

p l a n e ) p o s i t i o n i s w e l l k n o w n b u t n o z ( d e p t h ) i n f o r m a t i o n i sk n o w n . A c o r o l l a r y i s t h a t f o r a n o n - p a r a l l e l b e a m t h e r e w i l l a l s obe sm al l x , y er ro rs du e to paral lax. There are a lso hard edge effects aswe shall see be low. Compton scatter is a well understood and characterized effect(Evans1955) . T he scattered x-ray loses energy in the process, and for 30 and 60 keV x-rays, typical of thevalues of interest here, and a scattering angle of 145, the once-scattered x-rays have anenergy of 27 and 49 keV, respectively. The crossection for theprocess is given by theKlein-Nishina formula (Evans1955) and is nearly isotropic at these lowenergies.The justification for the safety of these machines rests on twoassertions. The first assertion is that the radiation used does not penetratebeyond the skin (for referenceson this subject see (Kaufman 2010)),

second, that the effective dose is no more than0.1 (or variously 0.05, 0.03,0.02 or 0.01) microSv (National Council on RadiationProtection andMeasurements2003a ,2003b).As we discuss below, at the kilovoltage (kVp)set ting of the uni ts in use, 50 kVpand 125 kVp , pe ne t ra t io n is a g iv ens ince t hese a re d i agnos t i c x - r a y un i t se t t i n gs ,and d iagnos isdepends on penetration through a subject.Regarding dose, except for thenumbers mentioned above, little information exists.The dose from these units is usuallyreferred to what is commonly known as SC 01-16(National Council on RadiationProtection and Measurements2003a) , w h i c h o n t h i s s u b j e c t i s a s u m m a r yo f S C 0 1 - 1 2 ( N a t i o n a l C o u n c i l o n R a d i a t i o n P r o t e c t i o na n d Measurements2003b), with both documents referring to ANSI/HPS StandardN43.17(Radiation Safety f or Personnel Securi ty Screening S ystems UsingX-ray or Gamma Radiation, ANSI/HPS Draft Standard N43.17). In all casesthese are documents that d e s c r i b e t h e r a t i o n a l e f o r a c c e p t a b l ee x p o s u r e l e v e l s a n d d i s c u s s m e t h o d s f o r measuringeffective dosein a variety of security devices. The calculation of dosein Sieverts (Sv) isbased on an involved use of tables and charts and on assumptionsaboutvolumes and organs being exposed-thus the importance of penetration.Theya l so i nc l ude some a l l owances f o r kV p se t t i ng . Raw da ta andt h e as su m p t io n s a nd recipes by whicheffective dosewas calculated for these units are not presented.To dist ingu ish our resu lt s fr omthe pub lished doses, we use the term exposure andw e g i v e i t i nt e r m s o f n a n o g r a y ( n G y ) s o t h a t i t i s c l e a r t h a t t h i si s a p h y s i c a l p a r a m e t e r c l e a r l y d e f i n e d a n d n o t


5/33

a d j u s t e d f o r e s t i m a t e s o f p e n e t r a t i o n , t h e k i l o v o l t a geu s e d i n t h e d e v i c e , a n d o t h e r f a c t o r s . T h e effective dose

publishedinc l udes fac to r s f o r the k i l ovo l tag e o f the un i t b e in gte st ed , an d, fo r in st an ce , is d i fferent for f ront and back exposure, andmay be three or six times smaller than theexposure based on energydeposited and the quali ty factor, which is 1 for x-rays.K n o w l e d g e o ft h e g e n e r a t o r a n d x - r a y t u b e u s e d , k V p s e t t i n g a n df i l t r a t i o n . locat ion with respect to col limat ion, col l imator crossect ion andlength, and scanningarea and speed, information of little use to a someoneintent on passing contraband,as we shall see below-, would permit anaccurate eva luat ion of dose . Of these , kVps e t t i n g s a n d f i l t r a t i o n ,s c a n n i n g a r e a a n d s c a n s p e e d a r e k n o w n f r o mv a r i o u s documents, including (National Counci l on Radiat ion Protect ion

and Measurements2003a ,2003b) a n d T S A o p e r a t i o n a lr e q u i r e m e n t s ( O p e r a t i o n a l R e q u i r e m e n t s Documents2006).Lacking the other parameters, a direct calculation is not possible. n this paper we show how an understanding of the performance parametersof the backscatter units can be obtained f rom what has been disclosedtogether with ananalysis of the published images.Anotherfactortobeconsideredissensitivitytocontraband.Twoviews ofthe subject arepresently obtained, one from the front and another from the back. Theavailableimages show hard-edged low atomic number Z objects such aswater-filled bottles orbricks

of explosive, and iron in the form of knives or guns. The low Z objectsaresometimes peripheral to the body, where the distortion of normal anatomyis clear, andthe high Z materials are always presented against the body as abackground.An important factor is safety, what the exposure levels mean interms of hazardsto exposed humans. This is a complex matter, since it needsto take into account dose to each organ, the skin, thyroid, fetus, etc. eachresponding differently, and the predisposition, genetic or disease -induced, ofthe exposed individual. The authorshave no expertise in the f ield of thebiologic effects of radiation, and we do not enter into the debate as towhether the levels calculated aresafeor not, or on r isk/benefit considerations. As a consequence, we presentexposure levels in terms of energyabsorbed by a gram of tissue, using thedefinition of the gray as 10,000 erg/gram for radiation with a quality factor of


6/33

1 (which is the case for x-rays and electrons).Exposure levels are given as afunction of depth in the tissue, this not being directlycomparable toeffective dose

in Sv presented in the published documents.Input parametersThe x-ray spectra are known from Refs. (National Council on RadiationProtectionand Measurements2003a ,2003b) and others, and are shown inFigure2for the two commercial units in use. The higher kilovoltage unitoperates at 125 kVp with 1.5mmAluminumfiltration,andthelowerkilovoltageunityat50kVpwith1mmofAluminumfiltration. The first is CT kilovoltage, the second is higher than that used in

Figure 2Published x-ray spectra for low and high kilovoltage units, from Refs.(National Council onRadiation Protection and Measurements2003a ,2003b) L.Kaufman, J.W. Carlson

mammography. We will refer to these as high and low kilovoltage units. Forcalculational purposes, the energy distribution of the bea m was approximatedbydiscrete steps (nine steps between 11 and 48 keV for low kilovoltage and12 steps between 17 and 116 keV for high kilovoltage, so that the K-lines oftungsten are wellrepresented). The beam is assumed to be a pencil beamwith no penumbra. As we willsee below, the beam dimensions affect spatialresolution but not exposure levels. Noknowledge of the beam fluence isneeded in this


7/33

analysis.Thesolidangleforthedetectorscanbereasonablydeducedfromphotosandfloorplansavailable on the web from the manufacturer of the Rapiscan1000(low kilovoltage). In our estimatesthesubjectisplacedsothat,fora25cm-thicksubject,thefrontis26cmfromthefront panel of the device (z direction). Thedetectors are assumed to be in side panels,against their front, and to be 176

cm in the vertical dimension (y), symmetrically placed,and extend from55 cm to29 cm and from +29 cm to +55 cm in the horizontal axis(x).ThedetectorsareassumedtobeNaI(Tl)with100%detectionefficiencyforanyphotonthat reaches their front. The images used in our analysis are shown inFigures3and4.RationaleAlthough the available images are obtained from websites, and consequentlycan bedegraded, the necessary information can be obtained from them, and

checked for consistency among various sources


8/33

Figure 3Backscatter imageat low kilovoltage. Pixeldimension is 2.1 mm


9/33

Figure 4High kilovoltageimage apparently unprocessedexcept for backgroundclipping.The annotations are bythe manufacturer

If we know the standard deviation of the counts in pixels in a uniformregionof the subject, then, on the reasonable assumption of Poissonstatistics, we knowhow many counts are needed to obtain the measuredstandard deviation. This ison the assumption that the image has not beensmoothed or sharpened or similarly processed. In such a case, anapproximate estimate can be obtained bymeasuring an edge spread functionand thus estimating a smoothing function andcorrecting for it.To measure


10/33

standard deviation, small and uniform regions of images wereevaluated byrecording the intensity value of each pixel in an area of interest andthestandard deviation (s.d.) calculated from these values. The signal tonoiseratio (S/N) is the mean of the counts in the region divided by the s.d.This ratiocan be distorted by non-linearity and by background clipping of

data, thus, weexamined the imagesbackground (air spaces) to assure ourselves that anyclipping present wasminimal.Together with the counts in a pixel we need to know its dimensions,so that we can obtain the counts per cm2. Note that if the beam is larger or smaller thanthe pixel, in so far as the pixel dimension itself is known, the countsarerepresentative of the exposure in that area as the beam scans thesubject. Thecounts/cm2are simply given by (10/pixel dimension in mm)

2. The dimension isobtained in our case by (1,760 mm/number of pixels in they direction). The 1.76 mheight is consistent within less than 5% withpublished cabinet size, with the height for a human, and with governmentrequirements as in Ref. (OperationalRequirements Documents2006).Image characterizationFigure3is a low kilovoltage image. It measures 19015.3 units in uniformregions,leading to a S/N ratio of 12 on a per image pixel basis. The verticaldimension isspanned by 835 pixels, yielding a pixel dimension of 2.1 mm.This necessitatesapproximately 144 counts from each pixel, translating to3,270 counts/cm2.Figure4shows a high kilovoltage image likely without spatial processing.Thisimage, smoothed, is widely distributed. The y extent is 696 pixels andthe pixeldimension 2.5 mm. The image does not appear to have a solid anglecorrection. Theedge transitions are sharp, which are enhanced by clipping.The background in thisimage is clearly clipped, we estimate that by about 14units. We measure, in a bright and uniform region 14521 units, this resultingin a S/N of 7. Assuming 14 units of clipping the S/N is approximately 8. Thisnecessitates from 49 to 64 counts in a pixel, translating to 784 to 1,024counts/cm2.Using Figure4as a reference, our exposure calculations will be referenced to55counts per 2.5 mm pixel. We will also show some results over a range ofcount/pixeland pixel dimension values. The conversion is simple: Exposureincreases linearlywith the number of counts per pixel and it decreases withthe square of the pixeldimension. Exposure increases as the square of theS/N in a pixel.Methods


11/33

The geant4 (Agostinelli et al.2003) Monte Carlo program was run withthegeometry as described above. In as much as possible, existing geant4models wereused as the basis of the simulation software. In particular, thehuman phantom modelof Guatelli et al. (Guatelli et al.2006) was used for theconstituent model of soft tissue and the physics components (low energy

Rayleigh scattering, photoelectriceffect, Compton scattering and gammaconversion) were included in the simulation.The general suitability of geant4for modeling low energy x -ray processed has beendescribed elsewhere (Fiejoand Hoff 2008). Initially, the performance of the programand data extractionwere studied at a beam energy of 60 keV, close to the averageenergy of thehigh kilovoltage unit. The data presented here were for high andlowkilovoltage spectra as described above. The beam had a 2.75 mm2.75mmdimension. We note again that beam dimension does not affect exposure.The runsconsisted of 50,000 incoming x-rays at high and low kilovoltage, forthe calculationof exposure and penetration. We found that to be able to makereasonable statementsabout sensitivity to explosives we needed much higher

statistics, and made runs of2.5 million x-rays at low kilovoltage and 1 million x -rays at high kilovoltagefor thecalculation of sensitivity.We performed six simulations at eachkilovoltage.Simulation 1: An x-ray beam impinging the center of a block oftissue of dimensions 40 cm40 cm (W in Figure1) and thickness along thebeam direction of 25 cm (T in Figure1).Simulation 2: The block of Simulation1 with an added block of tissue (A inFigure1) centrally located and 20 cm20cm dimension, 1 cmthickness. Except for a minor solid angle change due tobeing 1 cmcloser to the detector, the results of Simulations 1 and 2 shouldbethe same.Simulation 3: Same as Simulation 2, but the added block (A inFigure1) simulatesTATP explosive, or triacetone triperoxide, with density1.19 andcomposition C9H18O6.Simulation 4: Same as Simulation 3 but witha density of 1.2. The slightly different density was intended for checkingconsistency of results.Simulation 5: Same as Simulation 2, but the addedblock (A in Figure1) simulatesPETN explosive, or pentaerythritol tetranitrate,with density 1.8 andcomposition C5H8N4O12.Simulation 6: Same asSimulation 5 but with a density of 2.0. The slightly different density wasintended for checking consistency of results.The six runs will be referred toas Den0, Den1,...Den6, respectively followed byLow or High for the kVpsetting.The values extracted for this paper are: A. Energy deposition in eV inthe target in1 cm-thickness steps for an area of 1 cm1 cm, so tha tapproximating the tissuedensity by 1, the exposure in Gy is easily scaledfrom the energy deposited. B. The percent of events detected as a functionof their depth of origin in tissue. C. The x,yorigin in tissue of the detectedevents. D. The sensitivity to explosives under variousexposure values.Inaddition we performed imaging simulations of various objects.ResultsPenetration and exposureFigure5shows the exposure in nanoGy to obtain 55counts per 2.5 mm2.5 mm pixel for the high and low kilovoltage un its, usingonly tissue in the Den0 (subject surface at 26 cm from the detectors) and


12/33

Den1 (subject surface at 25 cm from thedetectors) configurations. In thefigure, the first cm of tissue encountered by the beam is listed as a depth of0 cm. For equal counts, the low kilovoltage unit deliversa four times higherentrance exposure, the exposures becoming equal at about 7 cmof depth.Figure6shows the same data summed to simulate exposure af ter twoviews

(the normal usage mode) of a 25 cm tissue thickness. It can be seen that forthe high kilovoltage case the exposure around the center drops to about 60%of thesurface exposure and is higher than that of the low kilovoltage unit. Fora given

Figure 5Exposure in nGy for tissue at high and low kilovoltage. The exposure isreferenced to 55 counts per pixel and a 2.52.5 mm2pixel dimension


13/33

number of counts, the average exposure through the 25 cm of tissue is 1.6timeshigher for the low kilovoltage unit. Thus, at 50 kVp, equal statisticsrequire higher entrance and average exposure than at 125 kVp. We note thatfewer than 2% of the photons in the low kilovoltage x -ray beam scatter intothe detector, and that for highkilovoltage less than 5% do. The detection

efficiency as a function of incoming x-rayenergy is shown in Figure7.Aswould be expected, very little of the beam scatters from the skin or even a 1cm below it. Figure8shows the deepest point in the tissue from whicheventsreach the detector. At low kilovoltage less than half the image countscome from thefirst cm of the subject, a significant amount come from within 5cm of depth, andsome as deep as 15 cm. At high kilovoltage, less than aquarter come from the first cm, a significant fraction come from the first 10cm, and some from as deep as20 cm. We note also that events assigned tothe pixel on which the beam rests alsocome from scattering at remote x,ylocations in the t issue (Figure9(highkilovoltage) and10(low kilovoltage)), asmany as 30% from farther than a cmfrom beam center at high kilovoltage

(Figure9).Beam penetration into tissue decreases the sensitivity to higherdensityexplosives. Consider that half of the events arise from below a cm ofdepth. A1 cm layer of low atomic number (Z) material, such as explosiveswith average Z of near 7, not that different from tissue, will stop moreincoming x-rays and increase


14/33


15/33

Figure 7Efficiency, defined as the number of counts recorded by the detectors as afraction of thenumber of x-ray inputs, shown as a function of input x -rayenergy

Figure 6SamedataasinFigure5but summed so as to simulate the exposure of a 25 cmthickness of tissue


16/33


17/33

scatter per cm of depth, but will also decrease the number of x -rays reachingdeeper into tissue and the backscattered photons from deep in, so thateffects in theexplosive layer and in t issue will be in opposite directions,reducing sensitivity.Penetration also provides an obvious explanation for air -

space shadows such as lungFigure 9Distance in tissue from beam center where x-rays forming the image have aninteraction, highkilovoltageFigure 8Deepest point of interaction in tissue of x-rays that form the backscatterimage

shadows. Consider 2 cm of tissue of density 1 followed by lung of density0.3. At high kilovoltage, the first 2 cm of depth contribute the same roughly


18/33

42% of events, but the other 58% are reduced by a factor of 0.3, so that theimage contains 60% asmany counts over lung than over solid tissue, thevalue that we measured in highkilovoltage units. At low kilovoltage the effectis less pronounced, and the imagesshow less conspicuous air spaces.Highdensity, high Z materials will scatter very few x- rays and when in front of

tissue will appear dark but with some background counts at high kilovoltage(about 3.5% of tissue signal).SensitivityFigures11and12show the sensitivityto explosives. Figure11has error bars that correspond to the statistics of theMonte Carlo calculation (approximately 50,000detected events for eachkilovoltage scaled to 55 counts at Den0). Figure12showsthe error barscorresponding to 55 counts. The statistics of the Monte Carlocalculation (s.d.0.45%) exceed the excursion of the different materials tested. TheDen0 andDen1 tissue only runs are within 1 s.d., as are the Den3 and Den4 pair(TATP, differing in density by less than 1%) and Den5 and Den6 runs(PETN,differing in density by 10%) are closely matched, raising confidence in

the statisticsof the calculation. The excursion for TATP amounts toapproximately 8% at lowFigure 10Distance in tissue from beam center where x-rays forming the image have aninteraction, lowkilovoltage


19/33


20/33

Figure 11Counts per pixel for the different configurations evaluated. Tissue alone at26 cm from thedetector plane is at the zero density point. At the density 1

point a cm of tissue is added to the tissue front.TATP with densities 1.19 and1.2, and PETN with density 1.78 and 2.0 are added to the configuration ofzero density. The statistics are those of the Monte Carlo calculation

Figure 12Same as Figure11with Poisson statistics of 55 counts

kilovoltage and it is4.5% for PETN. The excursions for the two explosives at highkilovoltage are

1.6% for TATP and1% for PETN.Table1shows standard deviations in percents under differentconditions.Recalling that the reference conditions are 55 counts per 2.5mm2.5 mm pixel,Table1shows the s.d. for an entrance exposure of 100nanoGy (the ANSI standardand current limit for these devices is 0.1 microSveffective dose) and a pixel of 6 mm6 mm. The s.d. at low kilovoltage is twice as large asthat at high kilovoltage,e.g., for fixed exposure and spatial resolution the lowkilovoltage images aregrainier, but its signal excursion, or difference from noadded object, for TATP isfour to f ive times larger. The excursions are abouthalf as large for PETN, a similar factor is calculated between the twovoltages.For S/N value of 12 and 20, to keep entrance dose below 100nanoGy, pixel area has to be larger than 6 and 20 mm2at high and low kilovoltage, respectively(Figure13).There exists some roomfor lowering of exposure results by a change in the input parameters used.The detector efficiency is assumed at 100%, even though for thekind oftechnology in use in this market, heavily influenced by number of units andbudget, and little influenced by competitiveness and performance, theefficiency mayeasily be 50%. Thus, the 100% efficiency used here is agenerous estimate. Thedetector outside dimensions are limited by cabinetdimensions. It is possible toincrease detection solid angle by narrowing ofthe inside distance between detectors,thereby increasing the horizontalangular variance of the x-ray beam, and suchnarrowing is included in theimaging simulations.Smoothing increases S/N with consequent penalty onspatial resolution. Figure14shows the same data as that of Figure4, butprocessed with smoothing and edgeenhancement, as evidenced by


21/33


22/33

Figure 13Scaling of entrance exposure at high and low kilovoltage. The range of 10 30 nGy (in Svunits) is widely quoted as the effective dose of the backscatter units (National Council on Rad iationProtection andMeasurements2003a ,2003b), while the shown values are entranceexposures

2% at the single pixel level. Given the 1 pixel -scale noise of 13.5%, thisrequires 46times more counts, and concomitant exposure. To reach Roses criterion the entranceexposure would need to be 2,070 nGy for a 2.5 mmpixel, or the pixel dimension17 mm for the exposure of 45 nGy. At anentrance exposure of 100 nGy the pixeldimension would need to be 11mm.Similarly, for TAPT at high kilovoltage, the SD is 2.0%, the 1 pixel levelnoise is13.5% at an entrance exposure level of 1.14 nGy. To meet Rose


23/33

s criterion the noiseshould be 0.5%, this necessitating 730 times more countsand exposure, leading toan entrance exposure of 8,320 nGy for a 2.5 mmpixel, or a pixel dimension of 68 mm at an entrance exposure of 11.4 nGy. Atan entrance exposure of 100 nGy the pixel dimension would need to be 23mm.For PETN the conditions are worse. At low kilovoltage the entrance

exposureneeded to meet Roses criterion is 6,210 nGy for a 2.5 mm pixel, or the pixeldimension 29 mm forthe exposure of 45 nGy. At high kilovoltage the entranceexposure is 41,000nGy for a 2.5 mm pixel, or the pixel dimension 150 mm for theexposure of11.4 nGy. At an entrance exposure of 100 nGy the pixel dimensionswouldneed to be 19 mm and 51 mm at low and high kilovoltage, respectively.It canbe argued that the eye is able to average signal over large areas, so that thepixels can be small and yet viewing conditions can be set for optimalaveraging. Thisis true when the noise in the image is stochastic at all spatialfrequencies. If there isstructure, e.g., variations such as those created by

anatomic variability and various


24/33

Figure 14Backscatter image at high kilovoltage. This image(same as Figure4) has abeensmoothed and edge enhanced.Pixel dimension is 2.1 mm

benign objects, as is clearly the case in the images, then the noise at anyone spatialfrequency is given by the quadratic sum of the stochastic andstructured noise (Shosa and Kaufman1981). Consider Figure3with a S/N of


25/33

12 on a per 1 pixel basis. If the eye were to average over an area of 10,000pixels (a diameter of 11 cm), the perceived S/N would be 1,200, sufficient forthe task of reliably detecting 18%variations. Nevertheless, over the spatial extent of 11 cm, the torso of

Figure3,which is far more featureless than the abdomen, shows a S/N of 5,whollyinadequate to the task, and not improved by reducing stochasticnoise.It is worth noting that Figure3, with a pixel dimension of 2.1 mm and aS/N of 12, this simulation requires an entrance exposure of 170 nGy, 70%higher than theANSI guideline.As we shall see below, these units depend onhard edge effects for the purpose of detection.

ImagingFor demonstration of image characteristics in x -ray backscattersystems, wesimulated an abdomen by a cylindrical (45 cm-tall in a 50 cmfield of view)uniform tissue elliptical phantom, 45 cm-wide and 25 cm-thick,the closest point of the front placed at a 26 cm distance from the plane of the

detector. For imaging,detector dimensions were changed from the aboveevaluations: While the height andoutside dimensions are limited by cabinetsizes, we assumed for imaging a smaller dead space between right and leftdetector, 18 cm rather than 58 cm. The effect of this change is to increasesensitivityat the centerby factors of 3.4 and 3.6 at highand low kilovoltage, respectively. Detectorperformance continued to be assumed at 100% for all photons reaching itssurface. The pixel size was 2.5 mm2.5 mm, andthe beam dimension 6mm6 mm. The display is for 5 mm5 mm pixels, with theaverage of the fourpixels forming it.Four examples were simulated. No added material, addedtissue, added TATP at density 1.2 and added PETN at density 2.0. Theseadded materials were configuredas a cone orpancake

of 20 cm-diameter, 1 cm-height at the center, tapering to zerothickness atthe border and conformal with theabdomen. The volume of thisadded material was approximately 160 cm 3, making the weight of PETNapproximately 320 g (note that theshoe bomberandChristmas bomber


26/33

werereported to carry 40 g and 80 g of PETN, respectively, both consideredsufficient to blow a hole in an aircraft fuselage.)We compared these imagesto those of a tissue brickof the same volume(15 cm10 cm, 1 cm-thickness, 150 g), the brick

geometry being typically used for demonstrations of backscatter detectioncapabilities.All tests included a water bottle of 3 cm-diameter and 5 cm-length, and an iron bar of 10 cm1 cm and 1 mm -thickness. For all images,the entrance exposure is10 nanoGy. Counts at this exposure are shown inTable2. It is important to note that the 10 nanoGy exposure images shownhere are a best case assumption. Detector efficiency is likely to be 50% ofthe value assumed, or even less at low kilovoltage,especially for multiple-scattered photons. The solid angle can be as much as 3.5times lower, and iscertainly less away from center, especially as departing center inthe verticaldirection, so the exposure could reach 70 nanoGy, close to 100 nGy


27/33


28/33

Figure 15Simulated images at high (a) and low (b) kilovoltage for 10 nGy entranceexposure of a 150 g brick of tissue (15101 cm). The center intensity is

approximately 4.5% higher than background because of edge effects. Notethat the brick is of the same composition as the underlyingabdomen. If thebrick were larger, its internal signal would be that of surrounding tissue

Figure 16Simulated images athigh kilovoltage for 10 nGy entrance exposure. datawere acquired in 2.5mm(200200matrix)pixelsandareaveragedanddisplayedwith5mmpixels.a Tissue only;b Added 1 cm-thick,20 cm-diameter pancake of 160 g of tissue;c Same as B, but with 190 g of TATP (density 1.2);

d Same as B, but with 320 g of PETN (density 2.0)

A characteristic of backscatter imaging is a hard edge effect akin to a highpassfilter with overshoot. Consider an x-ray beam scanning along a flatsurface from,say, left to right. Most of the x -rays scattered into the detectorcome from deep intothe subject, many not even from the beam position(Figures8,9, and10), and they leave through the surface. For a uniformsurface, neglecting solid angle changes, thescattered intensity is uniform. Asthe beam approaches a step that makes the materialthicker, deep-scatteredx-rays have more material to traverse if they scatter right rather than left. Atthe edge itself, half of the scattered x-rays have to traverse morematerial toexit, and there is a maximum drop in signal intensity. Consider now the beamscanning along the stepped material from right to left. Except for asmallchange in solid angle (proximity to detectors), the intensity is uniformand the sameas for the beam in the far left. As the beam approaches theedge, some deep-scattered photons now can exit through less material, andthe signal intensity increases,reaching a maximum at the edge. Thus, in ascan, there will be an undershoot at theedge on the lower surface, and anovershoot on the higher surface side. This effect will be unidirectional if thedetectors are not symmetrical with respect to the edgelocation.


29/33

Figure 17Simulated images at low kilovoltage for 10 nGy entrance exposure. data wereacquired in 2.5mm(200200matrix)pixelsandareaveragedanddisplayedwith5mmpixels.a Tissue only;b Added 1 cm-thick,20 cm-diameter pancake of 160 g of tissue;c Same as B, but with 190 g of TATP (density 1.2);d Same as B, but with 320 g of PETN (densit y 2.0)

Figure15shows the tissue brick at high and low kilovoltages. The brick isclearlyvisible since it is almost alledge


30/33

, given the beam spread shown in Figures9and10, and more so at high thanat low kilovoltage.Figure16shows the abdominal phantom at high kilovoltage.Figure16a is for the abdominal phantom with the bottle and iron alone.Figure16bincludes a pancake of tissue, which is not discernible. Figure16cand dare for TATP andPETN respectively. The TATP shows as a faint

increase in signal around itscenter, and the PETN also exhibits a faint darkring at its periphery. Given thefeatureless background and knowledge ofwhere to look, there are indications of the presence of these contrabandmaterials. Figure17is for the low kilovoltagecase. Again, knowing itspresence, the TATP is faintly visible as a signal increaseat the center, notthat different of what can be an anatomic feature (see Figures3,4,14and18).DiscussionThe publicly available information on backscatter airport scanners permits areasonable evaluation of their performance. The results indicate that theradiationused penetrates throughout the body, a not surprising fact to thosefamiliar withradiologic imaging at 50 and 125 kVp. Empirical support comes

from noting that theroutine diagnostic use of x-ray CT is practiced at 80 kVpto 120 kVp.Mammography, similarly needing enough x -rays leaving a breastof 10 cm or morethickness to make exquisite resolution and S/N images, ispracticed at well below50 kVp.


31/33

Figure15shows the tissue brick at high and low kilovoltages. The brick isclearlyvisible since it is almost alledge

, given the beam spread shown in Figures9and10, and more so at high thanat low kilovoltage.Figure16shows the abdominal phantom at high kilovoltage.Figure16a is for the abdominal phantom with the bottle and iron alone.Figure16bincludes a pancake of tissue, which is not discernible. Figure16cand dare for TATP andPETN respectively. The TATP shows as a faintincrease in signal around itscenter, and the PETN also exhibits a faint darkring at its periphery. Given thefeatureless background and knowledge ofwhere to look, there are indications of the presence of these contrabandmaterials. Figure17is for the low kilovoltagecase. Again, knowing itspresence, the TATP is faintly visible as a signal increaseat the center, notthat different of what can be an anatomic feature (see Figures3,4,14and18).

DiscussionThe publicly available information on backscatter airport scanners permits areasonable evaluation of their performance. The results indicate that theradiationused penetrates throughout the body, a not surprising fact to thosefamiliar withradiologic imaging at 50 and 125 kVp. Empirical support comesfrom noting that theroutine diagnostic use of x-ray CT is practiced at 80 kVpto 120 kVp.Mammography, similarly needing enough x -rays leaving a breastof 10 cm or morethickness to make exquisite resolution and S/N images, ispracticed at well below50 kVp.

The penetration not only distributes exposure throughout the body (thisaffectingthecalculationofefectivedose,whichcomprisesasumoverallorgans),buttendstodiffusethe effects caused by contraband materials. Images can be made at lowentranceexposures, but of very poor spatial resolution and S/N. Thecalculated signal excursionsat high kilovoltage are so small as to make itdoubtful that at any reasonable exposurelevels density differences will benoticeable unless the contraband is packed thickly andwith hard edges.Although the excursions are larger at low kilovoltage, they are stillsmall andin the noise of the devices operational limits. The eye is a goodsignalaverageratcertainspatialfrequencies,butitisdoubtfulthatanoperatorcanbetrainedtodetect these differences unless the material is hard -edged, not toolarge and regular-shaped. Anatomic features and benign objects addstructured noise that interferes withsignal averaging. Figure18shows awidely-distributed backscatter image. On the left is a complete view of hertorso, on the right, a section has been blacked out. While the breasts areeasily recognized at right, without some prior knowledge of the subject, it


32/33

would be hard to distinguish the increase of intensity in the superior part ofher breastsfrom the natural gradients of the image.It is very likely that alarge (1520 cm in diameter), irregularly-shaped, cm-thick pancake with beveled

edges, taped to the abdomen, would be invisible to thistechnology, ironically,because of its large volume, since it is easily confused withnormal anatomy.Thus, a third of a kilo of PETN, easily picked up in a competent pat down,would be missed by backscatterhigh technology. Forty grams of PETN, a purportedly dangerous amount, would fit in a 1.25mm-thick pancake of thedimensions simulated here and be virtually invisible.Packed in a compact mode,say, a 1 cm4 cm5 cm brick, it would bedetected.The images are very sensitive to the presence of large pieces of

high Z material, e.g., iron, but unless the spatial resolution is good, thinwires will be missed becauseof partial volume effects. It is also easy to seethat an object such as a wire or a box-cutter blade, taped to the side of thebody, or even a small gun in the same location,will be invisible. While thereare technical means to mildly increase the conspicuityof a thick object in air,they are ineffective for thin objects such as blades when theyare alignedclose to the beam direction

AcknowledgementThe authors are grateful for the valuable input provided by Professor PeterRez of the Physics Department, A rizona State University.ReferencesAgostinelli S, Allison J, Amako K et al (2003) Geant4

a simulation toolkit. Nuclear instruments andmethods in physics researchsection A: Accelerators, Spectrometers, Detectors and AssociatedEquipment506(3):250303. (http://geant4.cern.ch/ )Evans RD (1955) The atomic nucleus. McGraw-Hill Book Co., New York Fiejo PV, Hoff G (2008) Geant4 validation onmammography applications. Nuclear Science SymposiumConference RecordIEEE 34973498Guatelli S, Mascialino B, Pia MG, Pokorski W (2006) Geant4anthropomorphic pantoms. Nuclear ScienceSymposium Conference RecordIEEE 13591362Kaufman L (2010) Letter to the Editor. Re: airport full body scanners.JACR 7(8):655 National Council on Radiation Protection and Measurements


33/33

(2003a) Commentary No 16- Screening of humans for security purposesusing ionizing radiation scanning systems. Bethesda, M