R

D

WD

a

ARRA

KCDI

1

tciaretcd

te(pe

if

0d

European Journal of Radiology 72 (2009) 209–217

Contents lists available at ScienceDirect

European Journal of Radiology

journa l homepage: www.e lsev ier .com/ locate /e j rad

eview

ose and perceived image quality in chest radiography

outer J.H. Veldkamp ∗, Lucia J.M. Kroft, Jacob Geleijnsepartment of Radiology, C2S, Leiden University Medical Center, Albinusreef 2, 2333 ZA Leiden, The Netherlands

r t i c l e i n f o

rticle history:eceived 15 February 2009eceived in revised form 22 May 2009ccepted 22 May 2009

eywords:hest radiographyose

mage quality

a b s t r a c t

Chest radiography is the most commonly performed diagnostic X-ray examination. The radiation dose tothe patient for this examination is relatively low but because of its frequent use, the contribution to thecollective dose is considerable. Consequently, optimization of dose and image quality offers a challengingarea of research. In this article studies on dose reduction, different detector technologies, optimizationof image acquisition and new technical developments in image acquisition and post processing will bereviewed.

Studies indicate that dose reduction in PA chest images to at least 50% of commonly applied doselevels does not affect diagnosis in the lung fields; however, dose reduction in the mediastinum, upperabdomen and retrocardiac areas appears to directly deteriorate diagnosis. In addition to patient dose, alsothe design of the various digital detectors seems to have an effect on image quality. With respect to imageacquisition, studies showed that using a lower tube voltage improves visibility of anatomical structures

and lesions in digital chest radiographs but also increases the disturbing appearance of ribs.

New techniques that are currently being evaluated are dual energy, tomosynthesis, temporal subtrac-tion and rib suppression. These technologies may improve diagnostic chest X-ray further. They may forexample reduce the negative influence of over projection of ribs, referred to as anatomic noise. In chestX-ray this type of noise may be the dominating factor in the detection of nodules. In conclusion, opti-

pmen

mization and new develochest diseases.

. Introduction

Chest radiography is the most frequently performed diagnos-ic X-ray examination; it is of value for solving a wide range oflinical problems. X-ray images of the chest provide importantnformation for deciding upon further steps in the establishment ofdiagnosis, treatment and follow-up procedure. Chest radiography

emains the mainstay for diagnosis of many pulmonary diseases,ven despite recent developments in cross sectional imaging ofhe thorax, particularly computed tomography (CT). Advantages ofhest radiography over cross sectional imaging are lower cost, lowerose and speed of acquisition and diagnosis.

In a European Directive, the need for optimization of acquisi-ion techniques for X-ray imaging and limitation of patient dose is

stablished [1]. The effective dose related to a posterior–anteriorPA) radiographic chest image is about 0.02 mSv [2]. For com-arison, this is about 0.5% of a CT scan of the chest. Theffective dose related to the lateral chest image is approximately

∗ Corresponding author at: Department of Radiology, C2S, Leiden University Med-cal Center, PO Box 9600, 2300 RC Leiden, The Netherlands. Tel.: +31 71 5263689;ax: +31 71 5248256.

E-mail address: w.j.h.veldkamp@lumc.nl (W.J.H. Veldkamp).

720-048X/$ – see front matter © 2009 Elsevier Ireland Ltd. All rights reserved.oi:10.1016/j.ejrad.2009.05.039

ts will enlarge the value of chest X-ray as a mainstay in the diagnosis of

© 2009 Elsevier Ireland Ltd. All rights reserved.

a two times higher compared to the dose of a PA projection[2].

Although individual patient dose in chest radiography is rela-tively low, its contribution to the collective dose is significant dueto the frequent use of this examination. In the Netherlands, about athird of all diagnostic X-ray examinations are a chest X-ray [2]. Theassociated estimated contribution to the collective dose is about18% [2]. Similar figures are reported in other western countries[3,4]. Chest radiography may be implemented also in screening pro-grammes in some countries, this would have a substantial impacton the collective dose from chest X-rays.

The frequent use and diagnostic importance of chest X-ray makethat optimization of image quality and patient dose is an impor-tant area of research. Therefore, the relation between these aspectsis herewith discussed. Studies that investigated dose reductionand perceived image quality will also be discussed. The digitalchest X-ray acquisition technique will be reviewed and recent chestradiography technologies will be assessed, both with respect todiagnostic accuracy and radiation dose to the patient.

2. Dose and image quality in digital radiography

Dose in digital chest radiography mainly affects the noise inthe images. Noise in radiography can be defined as uncertainty or

2 Journa

itqts

riaedtote[wtbsoantDwn

titdcd

2

i

Fwds

10 W.J.H. Veldkamp et al. / European

mprecision of the recording of an image, i.e. unwanted stochas-ic fluctuations in the image. The most disturbing effect on imageuality (and thereby on diagnosis) is that noise can cover or reducehe visibility of certain structures. The loss of visibility is especiallyignificant for low contrast objects.

An important source of noise in X-ray images is related to theandom manner in which the photons are distributed within themage. In a uniform digital image, pixel values (that are associ-ted to the individual detector elements) will vary around theirxpected value. This type of noise is known as quantum noise. Theegree of the fluctuations is related to the exposure level: quan-um noise in a detector element is proportional to the square rootf the exposure level (noise expressed by the standard deviation ofhe signal). Here it should be noted that the signal in the detectorlement is proportional to the number of photons imparting on it5]. Therefore signal-to-noise ratio (SNR) and thereby image qualityill improve with higher exposure levels. For example, improving

he SNR by a factor 2 can only be obtained by increasing the dosey a factor 4 (assuming that quantum noise is the predominantource of noise). The resulting improvement of image quality willbviously have to be assessed clinically against the increased dosebsorbed by the patient. Apart from quantum noise other additionaloise sources must be considered in digital radiography, i.e. detec-or noise (for instance electronic noise) and anatomical noise [6–8].etector noise becomes more significant at low exposure levelshereas for higher exposure levels quantum noise and anatomic

oise will dominate in medical radiographs.Anatomic noise can be referred to as overlaying anatomic fea-

ures such as ribs, lung vessels, heart, mediastinum, and diaphragmn a chest radiograph. This occurs since chest radiography involveshe projection of a three-dimensional structure onto a two-imensional image [3]. It has been shown that anatomic noisean have an important negative effect on observer performance inetecting abnormalities especially in chest radiography [3,7,8].

.1. Detector technology

With the former film-screen systems, the range of patient dosen clinical practice was inherently limited by its sensitivity (speed

ig. 1. Digital versus conventional chest radiography. For film-screen systems, the opticalith digital techniques (lower row of images). Due to the better dynamic range at very low

ose, digital images are presented in similar gray values. However, increased quantum notructures.

l of Radiology 72 (2009) 209–217

class). Because of the small dynamic range, film-screen radiogra-phy images appear underexposed at low dose and overexposed athigher dose. With digital radiography underexposure or overexpo-sure is less likely to occur. This can be explained by wide dynamicrange of these detectors [9] (Fig. 1). However, as explained above,the selected dose level used will influence the quantum noise levelin the image and thereby the diagnostic potential. Low exposureswill still create images with clear appearance of gross anatomicalstructures but increased quantum noise will possibly hamper visu-alization of subtle anatomic and pathologic structures. This aspectof digital radiographic systems forms an extra challenge and oppor-tunity for optimizing patient dose and (perceived) image quality.

An additional aspect of digital radiography is that differencesin digital detector technology lead to differences in image qualityand dose. In the past decade in most western European hospitalsradiological film-screen (FS) radiography has been replaced by dig-ital radiography systems. Already in the early eighties, computedradiography (CR) with phosphor plate systems was introduced. Theradiation level received at each point of X-ray photons reaching thestorage phosphor screen, is stored in the local electron configu-ration by elevating the energy level of electrons (excitation) in thephosphor. After exposure, the photon energy stored in the phosphorplate is read out by a laser scanner and a digital image is obtained.

At first, the quality of these systems was moderate and the doseneeded for recording chest images was higher compared to theFS systems. Over the last 20 years the CR systems improved inboth dose requirements as in image quality. Recent advances in CRare more efficient collection of light by reading both sides of thescreen (dual-sided read CR) which results in an increased signal-to-noise ratio, line scan read out yields improved speed and theuse of needle-like phosphor allows for improved X-ray absorptionefficiency (a thicker phosphor) without loss of spatial resolution[10].

In the nineties, so-called direct radiography (DR) digital sys-

tems with a flat-panel detector (FPD) became available for chestradiography. In a short period, different digital radiography chestsystems were introduced for clinical use. In FPD’s the conversion ofthe latent X-ray image into a measurable signal most often occurs ina layer of Cesium Iodide (CsI-FPD), Gadolinium Oxisulphide (GOS-

density is directly related to the dose (see upper row of images). This is not the caseand very high doses, a clear image is shown. Or, in other words, irrespective of the

ise at low doses possibly hamperes visualization of subtle anatomic and pathologic

W.J.H. Veldkamp et al. / European Journal of Radiology 72 (2009) 209–217 211

F are sis ified t

Fficaiascidmjoiti

ctad(lfet

ewcfl

ig. 2. A radiograph of an anthropomorphic phantom (left image). Different lesionsimulated nodule; (C) simulated nodule. In the observer study the radiologist ident

PD) or Selenium (Se-FPD) in combination with a matrix of thinlm transistors (TFTs) from which the light (CsI, GOS) or electricalharge (Se), provoked in the layer, is read out and transformed intodigital image. Detectors that use a scintillator that converts X-ray

nto light, are called indirect conversion systems. The light is actu-lly transformed by photodiodes in the TFT-layer into an electricalignal. The detector material in direct conversion systems directlyonverts X-ray photons into an electrical charge. A selenium layers used for this purpose. A different technique uses charge-coupledevices (CCD) in combination with a scintillation layer. The chest isuch larger than currently available CCD chips. Solutions for pro-

ecting the chest on a CCD chip are the use of lenses or taperedptical fibers, at the cost of reduced dose efficiency and degraded

mage quality. A better solution is the uses of the so-called slot-scanechnique. This technique uses a linear array of small CCD detectorsn combination with a narrow X-ray beam that scans the chest.

Several studies have shown the advantages of digital systemsompared to FS [11–13], for instance the improved visibility ofhe mediastinal areas in the image. One of these studies used annthropomorphic phantom with simulated lesions to investigateetection of lesions in the chest for a digital CCD slot-scan systemFig. 2). It was found that with the digital system, the number ofesions observed in the mediastinum was almost twice the numberound with the FS system. For the lung lesions no significant differ-nce was found. The dose levels were roughly comparable betweenhe digital and the FS system (speed class 400).

In another study the diagnostic performance was compared foright different digital radiography chest systems [14]. The systemsere assessed for detection of simulated chest disease under clini-

al conditions. The following systems were regarded: four differentat-panel detector systems, two different charge-coupled device

mulated and attached on the phantom. (A) simulated interstitial linear disease; (B)he lesions and indicated the location of the detected lesions.

systems, one selenium-coated drum, and one storage phosphor sys-tem. Differences in diagnostic performance were found among theeight different digital chest systems in the configurations underwhich they were routinely applied in clinical practice. Interestingly,differences in detection rates could not be explained by dose (Fig. 3).The differences in detector design were given as explanation here.The DR systems significantly outperformed the single-sided readCR system with respect to image quality whereas the dose levelsused with the DR systems were lower. Furthermore, the resultssuggested that the scanning CCD or slot-scan technique gave bestdetection results. Due to the superior anti scatter properties of thesmall scanning detector (most scattered photons will go along it)a grid can be omitted, therefore this technique is associated withexcellent image quality at relatively low doses [13,15].

Several other researchers investigated or compared differentdigital systems. Better performance at low spatial frequencies forCsI-FPD systems compared to Se-FPD systems is found by stud-ies that use physical parameters to investigate image quality as afunction of spatial frequency. This is explained by the high atomicnumber and high density of CsI resulting in good capture of thelatent X-ray image. On the contrary, at the higher spatial frequenciesthe Se-based systems show better performance since these systemsshow less blurring of the image signal [16,17].

For indirect conversion detectors, apart from to the high atomicnumber and high density, the needle-like structure of CsI reducesthe spreading of light in the scintillator that allows the use of a

thicker layer with higher efficiency compared with unstructuredscintillators like GOS and regular CR systems [16–18].

An extensive overview of studies that investigated dose require-ments and image quality of various digital systems for chestradiography is given in a recent publication [19]. Fig. 4 gives an over-

212 W.J.H. Veldkamp et al. / European Journa

Fig. 3. Different digital systems with respect to dose and lesion detection were com-pared in a phantom study (Fig. 2). Dose for various digital systems varies (upperbar plot). Differences in lesion detection (lower bar plot) cannot be explained bydifferences in dose but by detector design.

Fig. 4. Common systems are given as a function of dose and image quality in clin-ical practice according to literature. The circles represent the uncertainty in theresults: the acquisition technique may vary in practice, the research methods dif-fer in literature and give an inherent limited view and finally, differences betweenmanufacturers may exist per detector type.

l of Radiology 72 (2009) 209–217

all impression of the results found in literature as partly discussedabove. Common systems are given as a function of dose and imagequality in clinical practice. The circles represent the uncertainty inthe results: the acquisition technique may vary in practice, differ-ent research methods are used in literature (each giving inherentlimited results), differences found between systems are not nec-essary unequivocal and finally, differences between manufacturersmay exist per detector type.

2.2. Diagnostic perception

Quantum noise and detector noise hamper detection of objectswith low contrast. Detectability of objects on a uniform backgroundis a function of object contrast and object size (detail). Detectabil-ity of objects can be improved by increasing radiation dose. It wasdemonstrated however that in radiographic chest images for detec-tion of lung nodules with sizes in the order of 10 mm, the projectedanatomy (anatomic noise) is far more disturbing than quantumnoise and system noise [8].

Anatomic noise can hinder detection through two processes:local influence, or “camouflaging,” and global influence, or “con-fusion” [6]. Local influence is for instance a rib that is projectedover a small nodule while global influence describes lesions thatfit very well in the global pattern of certain anatomical structures.Confusion is for instance a small nodule that may be interpreted asa tangentially projected vessel. Anatomical noise by the projectionof ribs is of particular concern for detection of lung nodules (i.e.,potential malignancies), because the ribs overlay about 75% of thearea of the lungs [3]. It was estimated that 50% of nodules mea-suring 6–10 mm are missed owing to superimposition of anatomylike ribs, heart, and mediastinal structures [20]. Interestingly, purelyon the basis of inherent contrast, a nodule as small as 3 mm indiameter should be visible on chest radiographs when anatomicnoise is not taken into account [3]. In general, it appears thatmany overlooked lung nodules in chest radiographs are visible inretrospect [21,22].

3. Study designs for investigating the effect of dosereduction on image quality

A number of methods for investigating the relationship betweendose and image quality have been developed. Objective mea-surements of physical characteristics, such as modulation-transferfunction, detective quantum efficiency or contrast-to-noise-ratio,and contrast-detail studies are often used. Alternatively, anthro-pomorphic phantom studies and clinical studies can be used forsubjective (quality rating) and objective (lesion detection) observerperformance studies.

The detective Quantum Efficiency (DQE) is a measure of thecombined effect of the noise and contrast performance of an imag-ing system, it is expressed as a function of object detail. Noise canbe expressed by the signal-to-noise ratio (SNR), contrast-to-noise-ratio (CNR) or by the noise power spectrum (NPS). An imagingsystem’s ability to render the contrast of an object as a function ofobject detail is traditionally expressed as its modulation-transferfunction (MTF). The combination of the functions NPS and MTFdetermines the above mentioned DQE. These objectives physicalmeasurements describe the systems technical imaging perfor-mance but it is still difficult to translate the outcome to the clinicalsituation that is far more complex than these measurements can

describe.

Contrast-detail studies can be used to determine dose levelsthat are related to a desired contrast-detail performance or canbe used to compare different systems or acquisition techniques.An advantage of contrast-detail studies over objective physical

W.J.H. Veldkamp et al. / European Journal of Radiology 72 (2009) 209–217 213

Fig. 5. Zoomed detail of a 0.162 mm pixel size matrix showing simulated low contrast nodular lesions located in a uniform object. Details of actual images are shown at thel pondt tual ds ose imh lower

mpdb

rtfmw

cilrtc

bsTia

fsrwodistsm

4g

sm

random order. Four radiologists ranked the quality of the corre-sponding images and rated diagnostic quality. The 100% referencedose was not recognized as the best quality image in nearly half ofthe cases (Fig. 7). Larger dose reductions to 25% and 12% were usu-ally recognized as third and fourth best quality. The preliminary

eft, whereas details of simulated images are shown at the right. The images correshe actual and simulated images appear quite similar. For 25 mAs the NPS for an acimulated reduced dose images are at the same level as their corresponding actual digher noise levels for higher spatial frequencies and slightly lower noise levels for

easurements is that contrast-detail performance includes theerformance of human observers. A limitation with both contrast-etail performance and physical measurements is that the anatomicackground is not taken into account.

Anthropomorphic phantoms better approximate the clinicaleality with respect to anatomic background. Some studies usehese phantoms in combination with artificial lesions that provideor a standard of reference in evaluating observer detection perfor-

ance. Unfortunately it is impossible to simulate appearance of thehole spectrum of diseases with artificial lesions.

An alternative is performing clinical dose experiments throughomputer simulation of the effect of dose reduction on image qual-ty (note that most often it is assumed unethical to perform clinicalow dose studies with real patients). A number of researcherseported on methods for simulating reduced dose images. Suchechniques have been applied in digital radiography [23,24] andomputed tomography [25–27].

A well-established method for reduced dose simulation that haseen previously described uses DQE and NPS at the original andimulated dose levels to create an image containing filtered noise.he method provides for simulated images containing noise that,

n terms of frequency content, agree very well with original imagest the same dose levels [23].

Another study described and validated a pragmatic techniqueor simulating reduced dose digital chest images, similar to noiseimulation techniques used for CT. This model includes using theaw pixel standard deviation as a measure of noise. Gaussian noiseith certain standard deviation is added to the original image to

btain a simulated reduced dose image with the desired pixel stan-ard deviation for each pixel [28]. After addition of noise the raw

mage was post processed in the standard way. Nodular lesions wereimulated in this study in combination with the low dose simula-ion (Fig. 5). Then, detection performance of simulated lesions inimulated and in real low dose images was investigated using thisethod by performing an observer model study [28].

. Dose reduction in chest radiography (how low can we

o?)

A number of studies have been performed to investigate the pos-ible clinical effects of dose reduction in chest radiography usingethods described above.

to 54 mAs (A) and 25 mAs(B) acquired with a scanning CCD chest system. Visually,ose and corresponding simulated dose image is shown at the right. The NPS of theages, although these NPS do not precisely fit as the simulated images show slightlyspatial frequencies.

The purpose of one low dose simulation study was to inves-tigate whether radiologists can rank the image quality of digitalradiographic images that correspond to different dose levels [29].Fig. 6 gives an impression of the appearance of different dose imagesby showing details. In this observer study, chest radiographs of20 patients (both PA and lateral images) with a variety of chestpathologies were used to simulate reduced dose levels correspond-ing to 50%, 25% and 12% of the original dose. For each patient theimages were displayed as hard copies on the film-viewing box in

Fig. 6. Detail images of the same patient with respect to 100%, 50%, 25% and 12% ofthe standard dose (respectively A–D). In the observer study, the entire images werepresented in a random order concerning the dose levels.

214 W.J.H. Veldkamp et al. / European Journal of Radiology 72 (2009) 209–217

F ers asq

ssdn

ar[uwdctermi(

dnpwpriprwrs

4

eicvappur

ig. 7. Ranking the images from 100% to the lower levels was difficult for the observuality image in nearly half of the cases for both PA and lateral images.

ubjective findings suggested therefore that a 50% dose reductioneems feasible in a variety of chest pathologies. However, furtherose reduction (more than 50%) clearly reduced the perceived diag-ostic quality.

In addition to their subjective study on dose reduction, the sameuthors also objectively investigated to what extent dose reductionesulted in decreased detection of simulated nodules in the lungs30]. Raw data from 20 normal clinical digital PA chest images weresed in this study. Low dose images were simulated and nodulesere attaches digitally. Hard copies were printed representing 100%ose and simulated patient doses of 50, 25, and 12%. These hard-opy images were reviewed by four radiologists. It was found thathe decrease in radiation dose from 100% to 50, 25, or 12% had noffect on lesion detection in the lungs. However, the decrease inadiation dose had a prominent effect on lesion detection in the

ediastinum. The detection performance of the four radiologistsn the mediastinum deteriorated with each dose reduction stepFig. 8).

A different study by other authors evaluated the influence ofifferent doses to the patient and peak kilovoltage settings on diag-ostic performance with respect to PA radiographic chest imageserformed with a FPD [31]. An anthropomorphic chest phantomas used in combination with simulated lesions attached on thehantom. The acquisition technique was varied when making chestadiographs of the phantom. Four radiologists assessed all of themages. For the lung fields only, no significant loss in diagnosticerformance could be demonstrated, even after a 50% reduction inadiation dose. This however did not count for the mediastinumhere the diagnostic performance deteriorated at this level of dose

eduction. This finding is in accordance with the conclusions by thetudy reported earlier.

.1. Acquisition technique

Diagnostic X-rays used for radiographs represent a certainnergy range: the X-ray spectrum. Contrast in a radiograph isnfluenced by the X-ray spectrum used. In general radiography aommon X-ray spectrum (of intermediate energies) usually pro-ides the best compromise between sufficient image quality and

cceptable patient radiation dose. However, in chest radiogra-hy such an intermediate X-ray spectrum would result in a veryronounced appearance of the ribs in the radiographs. Bone partic-larly effective attenuates X-rays at intermediate energies and theesulting bright appearance of the ribs in the radiographs would

is shown by the bar plots. The 100% reference dose was not recognized as the best

seriously hinder the evaluation of chest pathology. Historically, toovercome this problem, a compromise had to be made: chest radio-graphy is routinely performed using a relatively high energeticX-ray spectrum. This result in relatively ‘translucent’ representedribs but also results in less contrast in the soft tissues and particu-larly soft tissue lesions.

Some studies indicate that CR compared to FS can be used prop-erly in high-energy radiography of the chest providing equal imagequality at comparable effective dose to the patient [32]. Other stud-ies find improved SNRs or visibility scores when using lower tubevoltages under constant effective dose levels for CR and DR [33,34].Some recent studies however indicate that lower tube voltageschest radiography have as negative side effect prominent imag-ing of the ribs which may disturb the radiologist in his or herdetection [33,35]. Therefore in another study it is concluded that acompromise has to be made in tube voltage setting. It was demon-strated that an optimum X-ray spectrum for chest radiography withcesium iodide–amorphous silicon flat-panel detectors is obtainedwith 120 kVp and 0.2 mm of copper filtration [36]. In addition it wasconcluded from an anthropomorphic phantom study with humanobservers that standard PA chest radiographs should be acquiredwith 120 kVp (also routinely used) in terms of diagnostic perfor-mance and effective dose [31].

5. New developments in chest radiography

The strong current interest in, dual energy, tomosynthesis andtemporal subtraction have been made possible by the introductionof digital radiography.

5.1. Dual energy

An effective method to improve radiologic detection perfor-mance seems reduction or elimination of ribs, which has beenshown to be the main factor limiting the detection of subtle lungnodules as an early sign of lung cancer [3]. This can be achieved bydual energy chest radiography.

Dual energy comprehends weighted subtraction of low and

high-energy images and results in images representing bonestructures or images representing soft tissue. It was found thatdual energy added to standard PA chest radiography significantlyimproves the detection of small non-calcified pulmonary nodulesand the detection of calcified chest lesions [37].

W.J.H. Veldkamp et al. / European Journal of Radiology 72 (2009) 209–217 215

Fig. 8. Permission to reproduce these figures is given by the Radiological Society of North America (RSNA®). Original publication: Kroft LJM, Veldkamp WJH, Mertens BJA, etal. Detection of simulated nodules on clinical radiographs: dose reduction at digital posteroanterior chest radiography. Radiology 2006;241:392-398. (A–D) Clinical PA digitalradiographs of simulated chest nodules with (A and C) 100% radiation dose and with (B and D) simulated 12% dose. Images C and D show right lower part of radiographs Aand B, respectively. Diameter and attenuation of lung lesion were 3.24 and 4.00 cm, respectively (arrow in A and B). For mediastinal lesions, these were 1.62 and 2.25 cm,0.81 and 1.38 cm, 0.81 and 1.13 cm (arrow 1, 2, and 3, respectively). Note difference in noise levels between a and b and between C and D. Lung lesion is easily appreciatedin radiographs A and B, whereas subtle mediastinal lesions seem more difficult to appreciate at lower dose (D). (E and F) Graphs of number of simulated lesions detected in(A) lungs and (B) mediastinum at 100% dose and at reduced dose levels accumulated for all four observers. Note difference between lungs and mediastinum for effect of dosereduction.

2 Journa

tdRtawta

5

ianphFf

wwsd

na

5

etiidi

Fs

16 W.J.H. Veldkamp et al. / European

However the technique has some disadvantages as well: (1)he radiation dose is considerably higher compared to standardigital PA images which can be in the order of a factor 2–3, (2)educed signal-to-noise ratio in result images since the noise ofwo separated images adds up into the soft tissue or bone imagend contrast in the resulting images is reduced, (3) Additional hard-are is needed for making images at different energies in a short

ime interval; e.g. fast switching filters and fast detector read out,nd (4) extra images that may increase the workload.

.2. Temporal subtraction

Temporal subtraction comprehends subtraction of a currentmage and a prior image of the same patient. Standard PA imagesre used, so that no extra dose to the patient is needed. The tech-ique facilitates detecting pathologic changes over time and couldrovide diagnostic advantages. In the result images arisen nodulesidden by bones or the heart are potentially less likely to be missed.urthermore, infiltrative shadows should be better distinguishedrom breast tissue and pectoral muscles in the lung area.

Subtraction of current and prior images is not straightfor-ard because it is not likely that the two chest radiographsill have exactly matching projections. Temporal subtraction uses

ophisticated warping algorithms to eliminate the effect of theseifferences in projection.

The technique is currently only clinically used in Japan. Aumber of studies showed significant improvements in diagnosticccuracy with respect to nodule detection [38,39].

.3. Rib suppression techniques

Eliminating ribs has already shown to be effective with dualnergy suppression radiography despite the disadvantages related

o this technique that are hampers implementation in routine clin-cal practice. An alternative rib suppression method is based onmage processing techniques and has potential advantages overual energy radiography. Major advantage of such post process-

ng approach compared to a dual energy subtraction technique are:

ig. 9. Current implementation of the rib suppression method. The fully automatic methown at the left).

l of Radiology 72 (2009) 209–217

(1) the technique requires no additional radiation dose to patients,but uses only chest radiographs acquired with a standard digitalradiography system, (2) no specialized equipment for generatingdual energy X-ray exposures is required, and (3) noise levels arenot necessarily increased. Digital chest radiography provides suffi-cient large dynamic range and high bit depth to register accuratelythe entire range of radiation intensities in the latent PA chest image.This implies that visibility of lesions in the lungs should improveafter subtraction of intensities corresponding to ribs.

One paper describes a framework for rib suppression based onregression and applied on a pixel level. Radiographs with knownsoft tissue and bone images obtained by dual energy imaging wereused for training the framework [40]. Other authors presented morerecently a comparable approach [41]. They employed a massivetraining artificial neural network to create a rib suppressed imagealso using dual energy training data. A reduction of the contrast ofribs in chest radiographs is reported in both studies. A drawback ofthe two methods, likewise the dual energy suppression technique,is that the resulting images suffer from additional noise.

In Fig. 9 we present results for a rib supression technique com-parable to a method published earlier [42]. In this technique ribsand lung fields are segmented automatically using image process-ing techniques. The rib signal is estimated from image and modelinformation and subtracted from the rib (for each rib). By scaling thedifference value, the degree of suppression can be varied. Fig. 9 givesa result of the method used and shows the effect of rib suppression.

5.4. Tomosynthesis

Digital tomosynthesis is a method of producing coronal crosssection images using a digital detector and a chest X-ray systemwith a moving X-ray tube. In tomosynthesis, a series of low-doseexposures are made by moving the X-ray tube within a limited

angular range. The acquired images are processed into slices thatshow anatomical structures at different depths and angles. Thesecoronal cross section images have relatively high spatial resolu-tion in the image plane, but less resolution in the depth direction.Tomosynthesis can improve the visibility of pulmonary nodules by

hod suppressed the dorsal parts of the ribs in both lung fields (suppressed image

ourna

pos

asstnristli

6

racosipddtgiOa

R

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

W.J.H. Veldkamp et al. / European J

roducing cross section images without overprojection of ribs andverlying vasculature [43]. The actual effectiveness of tomosynthe-is is currently under study in a US national screening trial [43].

In tomosynthesis, the initial projection images show thenatomy from different orientations due to parallax. They arehifted and added together to render structures in one plane inharp focus. As a result objects in other planes are blurred out inhe resulting image. Deblurring algorithms are applied to elimi-ate the blurring of structures that are out-of-plane. A recent papereports that a typical set of acquisition parameters for chest imag-ng is 71 projection images, 20◦ total tube movement, and 5 mmpacing of reconstructed sections. The overall radiation exposureo the patient has been reported to be comparable to a former FSateral chest image. The entire acquisition sequence is completedn 11 s (which is within a single breath hold for most patients) [43].

. Conclusions and perspective

Digital chest radiography has many advantages over film-screenadiography, including improved diagnostic quality especially inreas of high attenuation (mediastinum, upper abdomen, retro-ardiac) and lower dose to the patient. There are a wide varietyf technically different digital systems, flat-panel detectors andlot-scanning systems are related to excellent image quality. Thentroduction of new techniques (tomosynthesis, dual energy, tem-oral subtraction and rib suppression) enables to further improveiagnostic accuracy. These techniques however must be furthereveloped and validated on a larger scale. A drawback may behe extra images that are added to the workload of the radiolo-ist. In conclusion, digital chest radiography is a time efficient andnexpensive investigation of low dose and good diagnostic quality.ptimizations and innovations may further improve the diagnosticccuracy.

eferences

[1] European Commission. Council Directive 97/43/EURATOM of 30 June 1997 onHealth Protection of Individuals Against the Dangers of Ionising Radiation inRelation to Medical Exposure, and Repealing Directive 84/466/EURATOM. Offi-cial Journal of the European Communities L-180/22 7-9-1997.

[2] Teeuwisse W, Geleijns J, Veldkamp W. An inter-hospital comparison of patientdose based on clinical indications. European Radiology 2007;17(7):1795–805.

[3] McAdams HP, Samei E, Dobbins III J, Tourassi GD, Ravin CE. Recent advances inchest radiography. Radiology 2006;241(3):663–83.

[4] Daffner R. Clinical Radiology, the Essentials. 2nd ed. Baltimore: Williams &Wilkins; 1999.

[5] Bushberg JT, Seibert JA, Leidholdt EM, Boone JM. Image quality. The Essen-tial Physics of Medical Imaging. 2nd ed. Philadelphia: Lippincott Williams &Wilkins; 2001. pp. 255–91 12-15-2001.

[6] Samei E, Flynn MJ, Peterson E, Eyler WR. Subtle lung nodules: influence of localanatomic variations on detection. Radiology 2003;228(1):76–84.

[7] Bath M, Hakansson M, Borjesson S, et al. Nodule detection in digital chestradiography: effect of anatomical noise. Radiation Protection Dosimetry2005;114(1–3):109–13.

[8] Hakansson M, Bath M, Borjesson S, Kheddache S, Johnsson AA, Mansson LG.Nodule detection in digital chest radiography: effect of system noise. RadiationProtection Dosimetry 2005;114(1–3):97–101.

[9] Veldkamp WJ, Kroft LJ, van Delft JP, Geleijns J. A technique for simulating theeffect of dose reduction on image quality in digital chest radiography. Journalof Digital Imaging 2008, 2-8-2008.

10] Cowen AR, Davies AG, Kengyelics SM. Advances in computed radiogra-phy systems and their physical imaging characteristics. Clinical Radiology2007;62(12):1132–41.

11] Aufrichtig R. Comparison of low contrast detectability between a digital amor-phous silicon and a screen-film based imaging system for thoracic radiography.Medical Physics 1999;26(7):1349–58.

12] Kroft LJM, Geleijns J, Mertens BJA, Veldkamp WJH, Zonderland HM, de Roos A.

Digital slot-scan charge-coupled device radiography versus amber and buckyscreen-film radiography for detection of simulated nodules and interstitial dis-ease in a chest phantom. Radiology 2004;231(1):156–63.

13] Veldkamp WJ, Kroft LJ, Mertens BJ, Geleijns J. Digital slot-scan charge-coupleddevice radiography versus amber and bucky screen-film radiography: compar-ison of image quality in a phantom study. Radiology 2005;235(3):857–66.

[

[

l of Radiology 72 (2009) 209–217 217

14] Kroft LJ, Veldkamp WJ, Mertens BJ, Boot MV, Geleijns J. Comparison of eightdifferent digital chest radiography systems: variation in detection of simulatedchest disease. AJR American Journal of Roentgenology 2005;185(2):339–46.

15] Samei E, Saunders RS, Lo JY, et al. Fundamental imaging characteristics of a slot-scan digital chest radiographic system. Medical Physics 2004;31(9):2687–98.

16] Borasi G, Nitrosi A, Ferrari P, Tassoni D. On site evaluation of three flat paneldetectors for digital radiography. Medical Physics 2003;30(7):1719–31.

[17] Samei E, Flynn MJ. An experimental comparison of detector perfor-mance for direct and indirect digital radiography systems. Medical Physics2003;30(4):608–22.

18] Kotter E, Langer M. Digital radiography with large-area flat-panel detectors.European Radiology 2002;12(10):2562–70.

19] Schaefer-Prokop C, Neitzel U, Venema HW, Uffmann M, Prokop M. Digital chestradiography: an update on modern technology, dose containment and controlof image quality. European Radiology 2008;18(9):1818–30.

20] Black C, Bagust A, Boland A, et al. The clinical effectiveness and cost-effectiveness of computed tomography screening for lung cancer: systematicreviews. Health Technology Assessment 2006;10(3), iii–x, 1.

21] Shah PK, Austin JHM, White CS, et al. Missed non-small cell lung cancer: radio-graphic findings of potentially resectable lesions evident only in retrospect.Radiology 2003;226(1):235–41.

22] Austin JHM, Romney BM, Goldsmith LS. Missed bronchogenic-carcinoma -radiographic findings in 27 patients with a potentially resectable lesion evidentin retrospect. Radiology 1992;182(1):115–22.

23] Bath M, Hakansson M, Tingberg A, Mansson LG. Method of simulating dosereduction for digital radiographic systems. Radiation Protection Dosimetry2005;114(1–3):253–9.

24] Treiber O, Wanninger F, Fuhr H, Panzer W, Regulla D, Winkler G. An adaptivealgorithm for the detection of microcalcifications in simulated low-dose mam-mography. Physics in Medicine and Biology 2003;48(4):449–66, 2-21-2003.

25] Frush DP, Slack CC, Hollingsworth CL, et al. Computer-simulated radiation dosereduction for abdominal multidetector CT of pediatric patients. American Jour-nal of Roentgenology 2002;179(5):1107–13.

26] Mayo JR, Whittall KP, Leung AN, et al. Simulated dose reduction in conventionalchest CT: validation study. Radiology 1997;202(2):453–7.

27] van Gelder RE, Venema HW, Serlie IWO, et al. CT Colonography at different radi-ation dose levels: feasibility of dose reduction. Radiology 2002;224(1):25–33.

28] Veldkamp W, Kroft L, van Delft J, Geleijns J. A Technique for simulating the effectof dose reduction on image quality in digital chest radiography. Jornal of DigitalImaging 2008, 2-8-2008.

29] Kroft LJM, Veldkamp WJH, Mertens BJA, Van Delft JPA, Geleijns J. Dose reduc-tion in digital chest radiography and perceived image quality. British Journal ofRadiology 2007;80(960):984–8.

30] Kroft LJ, Veldkamp WJ, Mertens BJ, van Delft JP, Geleijns J. Detection of simulatednodules on clinical radiographs: dose reduction at digital posteroanterior chestradiography. Radiology 2006;241(2):392–8.

31] Metz S, Damoser P, Hollweck R, et al. Chest radiography with a digital flat-paneldetector: experimental receiver operating characteristic analysis. Radiology2005;234(3):776–84.

32] Launders JH, Cowen AR. A comparison of the threshold detail detectabil-ity of a screen-film combination and computed radiology under conditionsrelevant to high-kvp chest radiography. Physics in Medicine and Biology1995;40(8):1393–8.

33] Chotas HG, Floyd CE, Dobbins JT, Ravin CE. Digital chest radiography with photo-stimulable storage phosphors - signal-to-noise ratio as a function of kilovoltagewith matched exposure risk. Radiology 1993;186(2):395–8.

34] Uffmann M, Neitzel U, Prokop M, et al. Flat-Panel-Detector chest radiography:effect of tube voltage on image quality. Radiology 2005;235(2):642–50.

35] Ullman G, Sandborg M, Dance DR, Hunt RA, Carlsson GA. Towards optimizationin digital chest radiography using monte carlo modelling. Physics in Medicineand Biology 2006;51(11):2729–43.

36] Dobbins JT, Samei E, Chotas HG, et al. Chest radiography: optimization of x-rayspectrum for cesium iodide-amorphous silicon flat-panel detector. Radiology2003;226(1):221–30.

37] Fischbach F, Freund T, Rottgen R, Engert U, Felix R, Ricke J. Dual-energy chestradiography with a flat-panel digital detector: revealing calcified chest abnor-malities. American Journal of Roentgenology 2003;181(6):1519–24.

38] Difazio MC, MacMahon H, Xu XW, et al. Digital chest radiography:effect of temporal subtraction images on detection accuracy. Radiology1997;202(2):447–52.

39] Uozumi T, Nakamura K, Watanabe H, Nakata H, Katsuragawa S, Doi K. ROC anal-ysis of detection of metastatic pulmonary nodules on digital chest radiographswith temporal subtraction. Academic Radiology 2001;8(9):871–8.

40] Loog M, van Ginneken B, Schilham AM. Filter Learning: Application to Suppres-sion of Bony Structures From Chest Radiographs. To appear in: Medical ImageAnalysis 2006.

41] Suzuki K, Abe H, MacMahon H, Doi K. Image-processing technique for sup-pressing ribs in chest radiographs by means of massive training artificial neural

network (MTANN). IEEE Transactions on Medical Imaging 2006;25(4):406–16.

42] Vogelsang F, Kohnen M, Mahlke J. Model-based analysis of chest radiographs.image processing; 2000. In: Proceedings of SPIE International Symposium onMedical Imaging. 2000.

43] Dobbins JT, McAdams HP, Godfrey DJ, Li CM. Digital tomosynthesis of the chest.Journal of Thoracic Imaging 2008;23(2):86–92.