Accepted Manuscript

Diagnostic X-ray sources –present and future

Rolf Behling, Florian Grüner

PII: S0168-9002(17)30587-9
DOI: http://dx.doi.org/10.1016/j.nima.2017.05.034
Reference: NIMA 59872

To appear in: Nuclear Inst. and Methods in Physics Research, A

Received date : 28 April 2017

Revised date : 23 May 2017
Accepted date : 24 May 2017

Please cite this article as: R. Behling, F. Grüner, Diagnostic X-ray sources –present and future,
Nuclear Inst. and Methods in Physics Research, A (2017),
http://dx.doi.org/10.1016/j.nima.2017.05.034

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form.
Please note that during the production process errors may be discovered which could affect the
content, and all legal disclaimers that apply to the journal pertain.
*Manuscript
Click here to view linked References

1 Diagnostic X-ray Sources – Present and Future

2 Rolf Behling a, Florian Grüner b

a
3 Philips Healthcare, Roentgenstrasse 24, 22335 Hamburg, Germany
b
4 Center for Free-Electron Laser Science and Universität of Hamburg, Luruper Chaussee 149, 22761 Hamburg, Germany

5 Abstract

6 This paper compares very different physical principles of X-ray production to spur ideation. Since more than 120

7 years, bremsstrahlung from X-ray tubes has been the workhorse of medical diagnostics. Generated by X-ray

8 segments comprised of X-ray tubes and high-voltage generators in the various medical systems, X-ray photons in

9 the spectral range between about 16 keV and 150 keV deliver information about anatomy and function of human

10 patients and in pre-clinical animal studies. Despite of strides to employ the wave nature of X-rays as phase sensitive

11 means, commercial diagnostic X-ray systems available until the time of writing still rely exclusively on measuring

12 the attenuation and scattering of X-rays by matter. Significant activities in research aim at building highly brilliant

13 short pulse X-ray sources, based on e.g. synchrotron radiation, free electron lasers and/or laser wake-field

14 acceleration of electrons followed by wiggling with magnetic structures or Thomson scattering in bunches of light.

15 While both approaches, non-brilliant and brilliant sources, have different scope of application, we speculate that a

16 combination may expand the efficacy in medical application. At this point, however, severe technical and

17 commercial difficulties hinder closing this gap. This article may inspire further development and spark innovation

18 in this important field.

19 Keywords: X-ray sources, X-ray tubes, medical imaging, synchrotrons, Thomson scattering, laser wakefield

20 acceleration.
21 1. Introduction

22 Medical imaging aims at producing visual and quantitative information about the anatomy of human patients

23 at lowest dose of ionizing radiation, see [1]. Bremsstrahlung sources yield a continuous spectrum, relatively large

24 average photon flux from millimeter-sized focal spots, incoherent and non-polarized radiation for exposure times

25 of between milliseconds and dozens of seconds. Figure 1 shows the interior of an X-ray tube in operation. Other

26 scientifically more advanced sources of X-rays are capable of delivering highly brilliant radiation, which is

27 highly monochromatic or at least features a narrow spectrum, is polarized, coherent and of pulse lengths of the

28 order of femtoseconds. It would be very beneficial if these characteristics in part or entirely could be transferred

29 to medical imaging in an affordable way. To name a few aspects: patient dose might be reduced by optimizing

30 the image contrast, scattered radiation and its quantum noise be differentiated from direct radiation, multiple

31 contrast agents and components of tissue be identified by their spectral characteristics. This paper discusses both:

32 affordable and compact bremsstrahlung sources on one hand and sophisticated highly brilliant sources on the

Figure 1 Rotating anode X-ray tube seconds after

exposure. One of the two electron emitting filament coils of
the thermionic cathode is still glowing. The X-ray focal spot,
where the electrons impact, is located underneath the anode
and not visible
33 other.

34
35 2. Current medical bremsstrahlung sources

36 2.1. Diagnostic X-ray sources and their application

37 Typically, high performance X-ray segments in diagnostic systems consist of an X-ray tube housing assembly,

38 comprising the vacuum electronics of an X-ray tube and means for radiation shielding, cooling, electrical

39 insulation and mechanical interfacing. Figure 2 discusses details of a typical high-performance tube for

40 angiography. See [2] for a comprehensive description.

41

42 Fig. 2 Cut view of a Philips (Hamburg, Germany) X-ray tube assembly for angiography and cardiology.

43

44 At first, a heated coil of tungsten wire in the negatively charged cathode, as shown glowing in Figure 1,

45 releases electrons into the vacuum space of the tube by thermionic emission. The cathode comprises focusing

46 electrodes, which may either be charged to the negative emitter potential or be biased by additional control

47 voltages for focusing, electron deflection or switching tube current and X-ray output off. An electric field

48 between anode and cathode accelerates the electrons. The total voltage defines the created spectrum of X-rays.

49 Alternative sources of fast electrons, like field emitters, e.g. comprised by carbon nanotubes, see e.g. [3], or laser-

50 driven acceleration of electrons in dielectric structures [4] have been proposed. But, still these methods do not

51 deliver the electron beams required for high performance rotating anode X-ray tubes.

52 The rotating anode, on the other hand, is typically charged positively (notably in bi-polar tubes). Recently,

53 single polar tubes are have gained popularity e.g. for high performance application in computed tomography

54 (CT), as they produce less off-focal radiation. In these tubes, only the cathode is charged (negatively). Single
55 polar tubes for mammography typically have the anode charged positively. Electrons hit the anode in a well-

56 defined focal spot area and generate X-rays, indicated in Figure 2 by the vertical arrow. Bremsstrahlung emerges

57 nearly isotropic into the entire hemisphere above the focal spot. X-rays for imaging are taken off at small angles

58 with respect to the anode shadow. Due to the angular isotropy of the intensity of bremsstrahlung, the ratio of

59 photon flux and optical focal spot size is maximal close to the anode shadow. As X-rays are generated about 1 to

60 3 micrometers below the surface, and as the anode roughens over time of operation, the intensity decreases

61 monotonically from about small angles of 10° above the anode shadow downward to 0° (so-called heel-effect).

62 The spectrum is tungsten-filtered, the more the smaller the take-off angle is. Tungsten cuts off low energy

63 photons on one hand and also high energy photons with energies beyond its k-edge at 69.5 keV. This

64 characteristic is of particular relevance for applications at high tube voltages, e.g. in computed tomography. The

65 focal spot is typically designed as a rectangular area oriented radially on the anode (Goetze focus). It appears

66 nearly square-shaped when seen from the center of the useful X-ray fan beam. The size of the projected focal spot

67 is minimal at low take-off angles. Due to their large area, notably when oriented orthogonal to the speed of

68 rotation, elongated focal spots provide enhanced thermal capacity, which is of great importance given the low

69 efficiency of conversion of electrical power to X-ray intensity of about 10-2. In rotating anode tubes heat

70 distributes by convection into the relatively wide focal spot track. This type of tubes delivers one to two orders of

71 magnitude higher photon flux than stationary anode tubes, see [2]. About 95% to 99% of the generated radiation

72 has to be scrapped in radiation shields. Lead layers of about 3 mm thickness extinguish undesired X-rays, which

73 emerge from the focal spot in all directions. Useful X-rays are being taken out only through the X-ray port. Using

74 tungsten targets, about 50% of the primary electrons are scattered back from the focal spot. Areas on the target,

75 hit by electrons at second impact, generate a haze of off-focal radiation. Fluorescence radiation from material in

76 the X-ray port adds. Target material, X-ray filters and the tube voltage define the beam quality. At a given tube

77 voltage, the tube current specifies the intensity of radiation produced. High performance angiography tubes, as

78 shown in the Figure 2 and Figure 3, allow for switching the electron beam off by applying a repelling voltage of

79 about 3 kV between electron emitter and control electrodes. High-performance CT tubes feature electrodes,

80 which slightly deflect the focal spot to suppress aliasing artifacts in the image.

81
82
83 Fig. 3 High performance angiography tubes in the production plant of Philips (Hamburg, Germany) waiting for exhaust, assembly into a
84 housing and testing.
85

86 The second important component of an X-ray segment is the source of electrical power. The high voltage

87 generator energizes the tube with instantaneous power of between several 100 watts for dental and surgery

88 imaging and more than 100 kilowatts in CT systems, see [2]. The generator also delivers control signals and

89 driving power for the tube rotor and heating the electron emitter. In addition, it controls X-ray output, X-ray

90 spectrum, and comprises service and safety features. High performance X-ray segments typically require a heat

91 exchanger, capable to dissipate average power of up to about six kilowatts to the ambient.

92

93 Types of bremsstrahlung sources differ by photon flux density, spectrum over time, shape and size of the X-

94 ray fan, mechanical interfacing, pulse sequence, and energy throughput per patient. A broad variety of sources

95 have been developed for the modalities CT, interventional cardiac and vascular X-ray (CV), interventional

96 surgical X-ray imaging, single exposure and fluoroscopic standard radiography, mobile X-ray, mammography,

97 and dental imaging. Figure 4 shows a comparison in terms of two of the key characteristics: spectrum defined by

98 the range of tube voltages used and electrical input energy per patient.

99
 Energy per
patient
(kWs)
1000 Computed tomography
Cardio / vascular
100
Mammo-
graphy
10
General radiography / mobile /
1 surgica / dental panoramic

0,1
0 50 100 150
Tube voltage (kV)

100

101 Fig. 4 Classification of X-ray systems by key characteristics of the X-ray source. Tube voltages are indicated on the abscissa. The ordinate

102 represents the energy per time period which is required for X-ray generation in most practical cases. For CT, mammography and general

103 radiography including mobile systems the integration period is defined as the X-ray-on-time it takes to image a single patient. The energy is

104 the product of tube voltage, tube current and exposure time. Cardio/vascular interventional imaging comprises a mixture of minute long series

105 of low-energy exposures interleaved with series of pulsed high power shots. The respective energy is stated in the chart for every five minutes

106 of an average interventional procedure.

107 The maximal tube voltage determines the design of high voltage insulation and with it tube size, cathode

108 design and radiation shields, and consequently the mass. The figures of energy throughput per patient, shown in

109 Figure 4, determine anode size and cooling means of the tube.

110 2.2. Computed tomography (CT)

111 As shown in Figure 4, clearly, CT is the most demanding modality, see e.g.[5,6]. The rapid development of

112 CT has spurred significant innovation in X-ray tube technology, see [7]. Typical use cases of this true 3D

113 imaging modality are soft tissue characterization in oncology, stroke management in neurologic imaging, trauma

114 triage in the emergency department, vascular (see Figure 5) and cardiac CT.

115
116
117 Fig. 5 Contrast enhanced and 3D rendered CT image of the renal vasculature (adapted from[2]).
118

119 The high performance X-ray tube shown in Figure 6 is rotatably mounted in a typical CT system. Figure 7

120 shows the example of a spectral sensitive CT system, Philips IQon®, see also [8]. Mathematically, an accurate

121 3D reconstruction of the attenuation pattern of a patient volume requires at least one line integral of the

122 distribution of the local attenuation coefficient in a so-called voxel inside the patient to be measured along each

123 coplanar direction through a “slice” of the patient. To acquire such a data set, the so-called sinogram, the X-ray

124 source has to rotate by at least 180° about the patient during acquisition of a “slice”. In a typical CT system, the

125 rotating source X-rays the patient from about 1200 angles about the center axis, in most cases making a full 360°

126 circle. Redundant data allow for improvement of the image quality. The high voltage generator on the rotating

127 gantry typically delivers between 30 kW and 120 kW of electric power with high voltage of between 70 kV and

128 150 kV. Slip rings transfer the electrical energy from the stationary unit to the rotating gantry. In order to scan

129 volumes beyond the coverage of the X-ray fan, which the detector can directly absorb, the couch of a system

130 translates the patient through the gantry bore while the X-ray tube is constantly spinning and radiating (helical

131 mode). An axial mode, where the patient table is standing still during scanning or is pushed forward in steps (step

132 and shoot), may also be used with multi-slice machines of sufficient detector size and patient coverage. Top tier

133 machines offer up to 16 cm tissue coverage with respect to the iso-center. This large coverage is beneficial for

134 cardiac CT. A single heartbeat may suffice for data acquisition. Another application of large coverage CT is

135 functional diagnostics, which delivers temporal sequences of scans, e.g. to monitor in-flow and washout of

136 contrast agents. Typically, CT scans require a total energy input to the tube of 800 kWs per patient or less.
 Central support plate 200 mm segmented all
metal anode

Pinched-off tubing Spiral groove bearing,

supported on both ends
Water in
Ceramics, magnetic field
Flat e- emitter bridge (rotor inside)

Cathode electron
drift path Water out
Ceramics insulator
-140 kV

e- Scattered electron trap

Focal spot on anode

(inside electron trap)
Double quadrupole
magnet lens &
Titanium X-ray port
dipoles
X - rays
Type: Philips iMRC tube

137

138 Figure 6 Cut view of the Philips iMRC® tube for the Brilliance iCT® and the spectral detection IQon® scanner families. (Adapted from [2]).

139

140 Fig. 7 Philips spectral detection CT system IQon®

141 In the past, single slice CT machines of the 1990’s demanded for much longer scan times and higher energy

142 input per patient than actual scanners. Utilization of photons was poor at the time due to limited detector size.

143 However, the historical trend to increasingly higher energy throughput per patient has reversed with the advent of

144 modern large coverage multi-slice machines, which offer rotation speeds of about 4 Hz and require higher

145 instantaneous tube power than legacy systems, whereas the energy throughput is reduced. Anode angles of 7

146 degrees suffice for systems with 4 cm coverage, which allows for relatively large focal spots. However, 16 cm

147 coverage demands for about 11 degrees. The focal spot must be physically shorter by the ratio of about 7/11 to

148 maintain spatial resolution in the centre of the image, which results in a higher desired power density in the focal

149 spot, by the inverse ratio, which the anode of the tube has to sustain. Priorities for heat management for computed

150 tomography developed from high energy per patient to high power density in the focal spot.
 Gantry rotation frequency / [Hz]
0,0 1,0 2,0 3,0 4,0 5,0
60 60
Centrifuga
Distance from
l accellera- 50 iso-center 50
tion / [1 g]
40 cm
40 50 cm 40
60 cm
30 30
70 cm

20 20

10 10

0 0
0 50 100 150 200 250 300
Gantry speed / r.p.m

151

152 Fig. 8 Centrifugal acceleration in a rotating CT gantry as a function of the angular velocity

153 The maximum speed of gantry revolution quadrupled in the past two decades. This resulted for current

154 systems in a 16-fold increase of the centrifugal forces to up to 40 times the gravitational acceleration g. Figure 8

155 relates gantry speed in terms of the angular frequency ω and centrifugal acceleration a of a revolving object in

156 units of g according to the relationship a = ω 2r, where r denotes the distance of the object from the iso-center of

157 rotation. Although the overall number of photons generated per patient went down significantly, the

158 instantaneous photon production per revolution has been more or less constant in order to maintain a sufficient

159 signal to noise ratio. Thus, the power rating of high-end CT systems raises proportional to the gantry speed.

160 Cardiac CT is particularly demanding in terms of tube power as scans may comprise only about half a single

161 gantry rotation in the optimal case. To avoid so-called aliasing artifacts high-end systems offer the ability to

162 toggle the position of the focal spot inside the X-ray tube electronically.

163

164 Beam hardening upon passing through tissue, bones and in some cases metal implants alters the integrated

165 attenuation along the line of projection, depending on the material scanned. When the spectrum changes upon

166 changing the perspective, an important condition for CT image reconstruction is compromised. One solution for

167 this severe issue of beam hardening is spectral CT, which also allows for material identification and proper

168 correction of the assumed spectrum, and tissue differentiation. Small amounts of contrast agents like iodine (for

169 angiography) or barium (for examination of the digestive tract) or gadolinium can e.g. be identified by their k-

170 edges of X-ray absorption at 33.2 keV (I), 37.4 keV (Ba) or 50.2 keV (Gd). (Along with discussing an alternative

171 synchrotron based X-ray source from the company Lyncean, Fremont, CA, USA, the source [9] provides more
172 details.) Varying the spectrum of primary radiation or measuring the ratio of hard and soft photons emerging

173 from the patient allows e.g. drawing iodine vessel maps even for complex structures in the skull, where iodine

174 appears interleaved with calcium. If more than two energy bins were available, one could even separate multiple

175 contrast media simultaneously. For an optimal dose-to-image contrast ratio, it would be ideal to have spectrally

176 tuneable monochromatic radiation generators, adapted to the diagnostic problem. There have been attempts to

177 monochromatize bremsstrahlung sources. Unfortunately, filter-based solutions as well as fluorescence-based X-

178 ray sources both suffer from low flux rates, and the spectrum is fixed. Using tungsten as a target, the yield of the

179 strongest k-alpha line of characteristic radiation is as low as 7% of the total intensity even at highest tube voltage.

180 Using thin targets may theoretically be a means to enhance the ratio of characteristic to continuous radiation.

181 Lorentz transformation results in forward enhancement of continuous bremsstrahlung, whereas characteristic

182 radiation emerges isotropic and can be taken off at the “back-side” of the impact of electrons in a very beneficial

183 way. But, the X-ray flux would suffer from the thin conversion layer. Alternative methods of spectral imaging, as

184 discussed below, are beneficial in this respect.

185

186 Source based X-ray spectral differentiation of image features in computed tomography, see [10], has been

187 commercialized first by General Electric (GE), Milwaukee, WI, USA and Siemens, Munich, Germany, on a

188 broad scale for computed tomography. See e.g. [11] for a comparison of the methods. Whilst GE Healthcare

189 modulates the tube voltage in cycles of several hundred microseconds [5], see also [12], Siemens uses a dual-

190 source–dual-detector [13] combination. Two pairs of independent tubes, X-ray filters, high voltage generators

191 and detectors rotate about the patient, sending orthogonal fans of X-rays of different spectra through the patient at

192 the same time.

193

194 The solution of Philips Healthcare, Andover, MA, in their Brilliance IQon® system, see Figure 7, is instead

195 based on spectral detection, see [8]. A stack of two layers of scintillator materials and photodiodes differentiates

196 photons of high or low energy after they have passed the patient and an anti-scatter grid on top of the detector.

197 Whilst low-energy photons are primarily absorbed in the inner scintillator, harder photons make it further out.

198 Unlike the aforementioned source-based methods, detector-based methods allows for fully accurate temporal
199 registration of the spectral information. Both energy channels are measured at exactly the same time and position.

200 Systems with direct conversion photon counting detectors exist as clinical prototypes, and also perform detection

201 based spectral differentiation, see [14]. The great benefit of detection based spectral CT imaging methods,

202 notably in times of growing use of metallic implants, is the intrinsic absence of beam hardening artifacts. In

203 addition, the spectral information is always at hand (“spectral always on”), no matter if the radiologist wants to

204 use it or not.

205

206

207 Fig. 9 Interventional cardiology image acquired during treatment.

208 2.3. Interventional cardiac and vascular imaging

209 The next demanding modality in the row is interventional cardiac and vascular imaging, which focuses on

210 conditions of the coronaries, see Figure 9, vasculature of the brain, peripheral vessels and interior vascular

211 diseases. Iodine based contrast agents help visualizing stenosis, bleedings, and aneurisms and generates images

212 with comparatively high contrast when an appropriate X-ray spectrum is used. The images are flat 2D projections

213 with vessels presenting as trees of iodinated partially opaque lines. Bleedings appear as shadows near highly

214 contrasting vessels. Subtraction techniques allow reduction of the “anatomic noise” from superimposed

215 structures, as shown in the neurological case of Figure 10. Typically, a movable C-arm carries an X-ray tube

216 assembly and a flat panel detector and allows for viewing the patient from different perspectives. Electrical

217 supply of the X-ray tube, high voltage cables, oil or water hoses for cooling fluid, and data cables run through the

218 stand. The high voltage generator is positioned in a remote engineering room. Studies are usually undertaken with
219 series of X-ray pulse patterns. Pulsed fluoroscopy is applied with focal spot size of less than a millimetre in each

220 dimension to monitor the placement of a catheter or other instruments. Pulse duration ranges from 2 ms to about

221 20 ms, frequencies between 7.5 Hz and 60 Hz with tube voltage between about 60 kV and 125 kV and tube

222 current of up to about 200 mA. Cine recording of the heart in motion and series of exposures to track the flow of

223 contrast die exploit the full tube power of up to about 100 kW and focal spots of about one millimetre in each

224 projected dimension for documentation of various heart phases and perspectives of the vascular structure.

225

226

227 Fig. 10 Interventional neurological vascular tree rendered as subtraction of images taken with and without

228 contrast agent (adapted from [2]).

229 2.4. Mammography

230 The special application mammography employs rather soft X-ray spectra and comparatively high image

231 resolution, primarily to identify micro-calcifications, which are characteristic for malign structures in the breast,

232 see [3] and [2]. In view of the possible image magnification, such spatial resolution requires small focal spots in

233 the range of 150 µm up to 0.5 mm width and projected length, and spectra below 50 keV photon energy. These

234 are often shaped by employing molybdenum targets - instead of tungsten - with a high content of characteristic

235 radiation and k-edge filters of the same material. Hard radiation would otherwise diminish contrast, notably when

236 used in film-based systems, see [15].

237 2.5. Other modalities

238 When sorted by tube voltage, general radiography, mobile X-ray, surgical C-arm systems turn out to be more

239 demanding than mammography. However, short exposure times and the low continuous X-ray flux in

240 fluoroscopic mode limit the overall energy per patient. Dental X-ray shares the same energy range, and is limited

241 to tube voltages below 75 kV.

242 3. General considerations for medical imaging

243 The potential hazards of typical dose rates of ionizing radiation for medical diagnostics have been subject to a

244 multitude of studies on a molecular as well as symptomatic level. Still, the debate has been ongoing without

245 univocal conclusion. As a consequence of this uncertainty, the ALARA principle (as low as reasonably

246 achievable) governs trade-offs between potential harm and diagnostic benefit in clinical practice. It is essential

247 when discussing brilliance of the X-ray source that the X-ray dose required for attenuation imaging a human

248 patient raises with the fourth power of the spatial resolution, see [16] and [6]. Phase contrast imaging enjoys a

249 smoother, but still quadratic relation, see [17] and a related discussion in [18] and [19], and [20,21]. Anyway,

250 spatial resolution has to be minimized to an acceptable level which just suffices for diagnostic outcome. Physical

251 methods of X-ray generation which aim at higher resolution must therefore deliver clinical proof of a largely

252 enhanced diagnostic value. At this point in time greater spatial resolution than delivered by state-of-the art

253 systems is very hard to justify clinically. Instead of brilliance, the dominant key performance factor is X-ray

254 photon production per characteristic time, which is e.g. the duration in which a human heart is at rest. That is the

255 basic reason, why bremsstrahlung sources with low brilliance and high photon flux rates compared with advanced

256 accelerator-based systems prevail. Figure 11 shows the production capacity of a modern CT tube. Filtered by a

257 realistic X-ray filter of 0.6 mm titanium, a modern tube is capable to deliver a photon flux in the order of about 2

258 x 1016 photons per second into the hemisphere above the focal spot. This flux can be maintained for a typical scan

259 time of at least 4 seconds, which yields an integrated photon number of about 10 17 photons per shot. The portion

260 directed into the used beam will depend on the system geometry and amounts typically to in the order of 1014 to

261 1015 photons per second. Alternative methods still suffer from deficiencies in this respect. For example, as a
262 reference, the Lyncean compact synchrotron source mentioned above, see [9], is reported to deliver 1.4 x 10 10

263 photons per second. For comparison with other X-ray sources Figure 12 shows photon flux and brilliance of a

264 high performance tube in units discussed in section 6.

Photons produced in
4 seconds of a CT Tube current
scan (filtered) Capability of a modern CT tube (mA)
10 18
1,0E+18 10 A
10000

10 17
1,0E+17 10A17
1
1000

Tube current (mA)

10 16
1,0E+16
Total # of photons produced in
0.1 A
100

4 s (tube filter passed)

15
1,0E+15 0.01 A
10
10
40 60 80 100 120 140
Tube voltage (kV)

265

266 Fig. 11 Photon production capacity of a modern CT bremsstrahlung X-ray tube for medical imaging

267

Brilliance ( #photons/(s * mrad2 Photon flux (# of photons / (s * keV)

* mm2 * 0.1%bandwidth) ) 140kV, W, 7°, 0.6 mm Ti filter
Bremsstrahlung from a modern CT tube
10 11
1,00E+11
10 17
1,00E+17

10 10
1,00E+10 10 16
1,00E+16

10 9
1,00E+09
10 15
1,00E+15

10 8
1,00E+08 10 14
1,00E+14

10 7
1,00E+07 10 13
1,00E+13

10 6
1,00E+06 10 12
1,00E+12

10 5
1,00E+05 10 10
1,00E+11

10 4
1,00E+04 10 9
1,00E+10

10 30 50 70 90 110 130 150

Photon energy (keV)
268

269 Fig. 12 Photon flux density and brilliance of a modern CT tube

270 4. Advanced applications in medical imaging

271 4.1. Differential phase contrast imaging / dark field imaging

272 Differential phase contrast imaging can display anatomic details in an unprecedented way, see [22], and

273 promises potential benefits when fine resolution is requested, see [20,23]. Comparing interferometer based and

274 propagation-based methods from a dose perspective the propagation method reveals conceptual benefits for

275 imaging of small objects. For structure sizes below about 30 µm the propagation-based method is superior even

276 to a monochromatic grating-based system. The propagation based method requires sources of high brilliance with
277 a high spatial coherence, i.e. focal spots of a diameter of about 10 µm and highest possible power rating. Such a

278 source is commercially available from the company Excillum, Kista, Sweden. A liquid metal jet serves as the

279 anode to generate bremsstrahlung. Instead, [22] promotes the grating-based method with a Talbot-Lau

280 interferometer, which seems more appropriate for objects of the size of a human breast and up. Conventional X-

281 ray tubes can produce spatial and temporal coherence of a superimposition of a multitude of beams using a

282 “coherence-“ slot grating upstream of the patient. The use of X-ray tubes with non-coherent polychromatic

283 radiation makes the method potentially applicable outside of synchrotron facilities. The downside of the grating

284 solution is photon flux reduction by the first coherence grating, and the necessity of an analyser grating

285 downstream of the patient which absorbs half of the radiation after it has already passed the patient. In addition to

286 a phase shift and attenuation image an X-ray dark-field image is generated as well. Small scattering objects in the

287 beam, which destroy coherence, can be identified akin visual dark-field imaging. This may deliver additional

288 signals from microscopic objects of sizes far below the ordinary limits of spatial resolution

289 4.2. Former attempts for improvement

290 History of bremsstrahlung is full of great achievements, but also ideas which failed to hit the market place (so

291 far). These include, e.g. a non-revolving circular X-ray tube for a so-called fourth generation CT system (issue:

292 scattered radiation), stationary CT tubes with carbon nanotube (CNT) field emission cathodes (issue: scattered

293 radiation at large coverage), inverse geometry systems in CT (issue: low flux) and in angiography (issue: slowed

294 clinical workflow during an interventional procedure), fluorescence tubes (issue: low flux), and others.

295 5. Prevailing issues of current diagnostic X-ray sources

296 a) The conversion factor of electrical energy to X-ray energy is in the order 10-4. Costs for electrical

297 infrastructure and devices for the conversion of electrical energy to X-rays are of the order up to several hundred

298 thousand dollars.

299 b) Thermal issues associated with the available target technology limits the spatial resolution of imaging

300 systems and the speed of acquisition.

301 c) Electronic space charge in the cathode limits tube currents, available X-ray flux, and tube life.
302 d). Bremsstrahlung is polychromatic, limiting the ratio of patient dose to image contrast resolution.

303 e) Poor spectral differentiation. The spectrum of the bremsstrahlung from conventional X-ray tubes is only

304 slowly tuneable by tube voltage in hundreds of microseconds.

305 f) Using bremsstrahlung excited fluorescence radiation results in a fixed line spectrum and typically

306 photon output which is multiple orders of magnitude too low for fast imaging, see [24].

307 g) The spatial X-ray flux profile within a beam can only be modulated with low spatial frequency.

308 h) X-ray lenses with wide enough capturing angles for photons of relevant energies (16 keV to 150 keVdo

309 not exist.

310 i) The patient skin dose at the entrance of the conic X-ray beam is of concern; a minimal distance to the

311 source is thus required and with it a minimal size of the X-ray system.

312 j) Bremsstrahlung from an ordinary X-ray tube target emerges in all directions; this requires massive lead

313 shields.

314 k) In addition to ionizing radiation, potential lethal voltages close to patient and staff and electrical as well

315 as hazardous mechanical and thermal energies have to be managed.

316 l) Bremsstrahlung from ordinary X-ray tubes is substantially incoherent. Differential phase contrast

317 imaging and dark-field imaging are possible, but require extra effort.

318 m). Bremsstrahlung from reflection targets is non-polarized. Potential information in this respect is absent.

319 n) The temporal definition bremsstrahlung is limited and would not allow e.g. for time-of-flight techniques

320 to differentiate between direct and scattered photons. Highly efficient detectors with suitable temporal response

321 are missing.

322 o) X-ray tubes wear out. Costs of ownership of a tube are substantial in relation to overall system costs.
323 6. Laser-driven brilliant X-ray sources

324 6.1 Basic physics of all-optical X-ray sources

325 The issues discussed above in chapter 5 as well as the emerging field of laser-driven X-ray sources have triggered

326 new thinking towards clinic-/laboratory-sized X-ray sources with unprecedented brilliance levels. The definition

327 of (average/peak) brilliance B is given by: B = number of photons  s-1 / (mm2  mrad2  0.1%) [25]. The average

328 brilliance refers to continuous sources delivering the photons constantly over time, while peak brilliance refers to

329 the number of photons per shot. The latter definition stems from pulsed sources, e.g. synchrotrons and, especially,

330 Free-Electron Lasers (FELs). However, for medical imaging the typical length of FEL-pulses on the order of

331 femtoseconds (10-15 sec) is so short (and the corresponding dose so high) that for most envisaged medical

332 applications only the average brilliance is discussed here. While conventional X-ray sources can deliver large

333 number of photons (over at least 1 second), their intrinsic problem is the very large bandwidth, divergence, and –

334 to a lesser extent – the focal spot size. Therefore, the primary focus on the search for advanced brilliance is to

335 reduce the bandwidth and divergence drastically. To illustrate this point, we discuss the imaging modality of X-

336 ray fluorescence, where the spatial resolution is maximal when using pencil beams [26,27]. If one requires a

337 pencil beam with smallest source size and a divergence at or below 1 mrad, and, in addition, a bandwidth of only

338 10% FWHM, the flux of X-ray tubes is at least three orders of magnitude too low. Therefore, there is a request

339 for new thinking about brilliant X-ray sources with intrinsically small source size, photon beam divergence, and

340 bandwidth.

341 If one aims at an intrinsically large brilliance, it is clear that the stochastic process of bremsstrahlung needs to be

342 replaced by a radiative process over which there is some degree of control. For instance, the photon beam

343 divergence is confined in the forward direction when relativistic electrons are used instead of the non-relativistic

344 few hundred keV electrons in X-ray tubes. The reason is that relativistic electrons emit their radiation within the

345 well-known opening angle of synchrotron radiation. For 100 MeV electrons, this angle amounts to only about 5

346 mrad, drastically smaller than the bremsstrahlung’s 2π solid angle.

347 Furthermore, the bandwidth can also be confined if there is control over the electron trajectory. For instance, the

348 so-called undulator radiation is emitted by electrons undergoing a sinusoidal orbit within a periodic, alternating

349 magnetic field. These insertion devices in straight sections of storage rings (synchrotrons) can produce X-rays

350 with a brilliance exceeding 1021 ph/(s mm2 mrad2 0.1% BW). BW means the energetic bandwidth in % which is

351 considered for counting the relevant photons. The main reason for such large brilliances is the intrinsically low

352 level of divergence and bandwidth, as well as small source sizes and large photon fluxes. However, for X-ray

353 photon energies on the order of 100 keV, such synchrotrons are operated with GeV-scale electrons (which, on

354 one hand immediately explains the ultra-small beam divergence, but on the other hand requires large-scale

355 infrastructures).

356 The next step towards compact brilliant X-ray sources suitable for installation in clinics is to replace the

357 permanent magnets of undulators by a laser pulse. Since the wavelength of (high-power) lasers are typically four

358 orders of magnitude shorter than conventional undulators, the electron energy necessary for the same photon

359 energy is favourably reduced from the GeV-scale to few ten MeV. The effect is based on Thomson (or, more

360 generally, Inverse Compton) scattering: if a laser photon is scattered with a relativistic electron, its wavelength is

361 upshifted from eV-scale up to the order of 100 keV and beyond [28]. The electron trajectory within a colliding

362 laser pulse is equivalent to the sinusoidal motion within a fixed permanent magnet undulator.

363 In order to realize an all-optical, hence compact, X-ray source, also the electron accelerator has to be driven by a

364 laser pulse. Since the seminal paper from 1979 [29] and the experimental breakthrough [30] , the field of the so-

365 called laser-plasma wakefield accelerators is emerging. The time span between the concept and first experimental

366 validation was determined by the lack of the required high-power, short-pulse lasers, that is, lasers with peak

367 powers on the order of few hundred TW and pulse length on the order of few ten femtoseconds.

368 For 100 MeV electrons, the acceleration distance in such a laser-plasma wakefield accelerator is on the order of

369 less than 1 cm. Therefore, together with the lab-sized footprint of the high-power laser, this Thomson X-ray

370 source is compact enough to be installed in clinical environments. The main reason for this short acceleration

371 distance is to be found in the extremely high electric field gradient within a plasma wakefield, which can reach

372 values of up to 1 TV/m. Such a wakefield is formed and trails the laser pulse through the plasma, if the laser

373 provides an intensity level called relativistic intensity. This is the case if the ponderomotive force [31] of the laser
374 field is so strong that an electron at rest reaches relativistic energies after one laser cycle. The laser then pushes

375 plasma electrons aside, which return to the laser propagation axis, hence forming a “bubble” shaped wakefield,

376 that is, a spherical electron-free cavity behind the laser pulse. Due to the corresponding charge separation, such

377 high field gradients arise and accelerate plasma electrons, which have been scattered into the wakefield.

378 A small portion of this accelerating laser pulse is used for the Thomson scattering pulse to be collided with the

379 wakefield electrons after they exit the plasma into the vacuum. Such a Thomson source is well tunable from laser

380 shot to shot by controlling the final electron energy via adjusting the accelerating laser pulse.

381 An alternative to such an all-optical Thomson source is a so-called betatron source [32]. Basically, every plasma

382 wakefield accelerator also generates betatron radiation, which is fully equivalent to the undulator radiation,

383 whereby the sinusoidal trajectory is imprinted on the electrons accelerated in the wakefield by the (focusing)

384 transverse wakefields. However, there are two typical drawbacks for betatron sources when it comes to medical

385 imaging: first, their (mean) photon energy is relatively little (rather below the envisaged 100 keV scale) and its

386 spectrum is intrinsically broad, because the electron trajectories have relatively large amplitudes (causing the

387 emission of higher harmonics) and a large variation. Due to the large electron trajectory amplitudes, the resulting

388 photon beam divergence is significantly larger than in case of Thomson sources.

389

390 6.2 The road towards laser-driven Thomson sources

391 While the perspectives for such all-optical, compact, and brilliant X-ray sources are manifold and promising,

392 there are still key hurdles to be taken. One has directly to do with the advantage of small bandwidths: for a

393 Thomson source, this is given only if the energy spread of the plasma wakefield accelerated electrons is

394 sufficiently small. The above described “bubble regime” typically delivers large energy spreads, because the

395 wakefield trapping of plasma electrons is also a stochastic process causing different final energies for electrons

396 injected at different times. This problem is approached by different ways and nowadays the energy spread seems

397 small enough [33], however still at cost of low electron charge, where one has to note that the number of photons

398 emitted scales linearly with the number of electrons.

399 From today’s view the most challenging issue is the photon flux, although it is already more than one order of

400 magnitude larger per single shot than the continuous flux per second from modern X-ray tubes. For instance, in
401 order to fulfil the requirements of X-ray fluorescence imaging with pencil beams, the repetition rate of the high-

402 power laser must be upgraded from the typical 0.1 – 1 Hz up to kHz-level. Such high-repetition rated laser

403 systems are under development [34], but as of today, they are still far away from delivering sufficient laser pulse

404 intensities required for the Thomson sources described above. In the next years prototypes delivering laser pulses

405 energies on the 1J-scale are expected and in the meanwhile the Thomson sources will be advanced by using low-

406 repetition rate laser systems for demonstrator experiments.

407

408 Medical imaging with laser-wakefield driven sources, be it based on Compton scattering or other means of

409 wiggling of the electron trajectory, would require scanning techniques to image at least entire organs of a patient,

410 extended lesions, portions of the vascular systems and adjacent tissue. There have been attempts to modulate the

411 interaction of the laser beam with the plasma cell for this purpose or to employ the angular spread of photons

412 after Thomson scattering. The direction of the input laser can easily be altered in laser-wakefield driven sources.

413 This would allow for scanning. The great benefit of a narrow and tuneable spectrum would simplify X-ray

414 fluorescence diagnostics and other spectral sensitive methods like k-edge imaging. Widening the divergence of

415 the X-ray beam after Thomson scattering is another option, which is currently pursued. But, the solution space

416 seems limited in this respect. On top, the angular dependency of the photon energy imposes a decent degree of

417 complication.

418

419 Still, medical X-ray diagnostics has been waiting for compelling solutions beyond the capabilities of X-ray tubes

420 with all their benefits and pitfalls, see [35].

421 References

422 [1] J. Beutel, H.L. Kundel, R. van Metter, Hanbook of Medical Imaging, (2000).

423 [2] R. Behling, Modern Diagnostic X-Ray Sources - Technology-Manufacturing-Reliability, 1st ed., CRC

424 Press - Taylor and Francis Group, LLC, Boca Raton, FL, USA, 2016.

425 [3] E. Gidcumb, B. Gao, J. Shan, C. Inscoe, J. Lu, O. Zhou, Carbon nanotube electron field emitters for x-ray
426 imaging of human breast cancer., Nanotechnology. 25 (2014) 245704. doi:10.1088/0957-

427 4484/25/24/245704.

428 [4] E. a. Peralta, K. Soong, R.J. England, E.R. Colby, Z. Wu, B. Montazeri, C. McGuinness, J. McNeur, K.J.

429 Leedle, D. Walz, E.B. Sozer, B. Cowan, B. Schwartz, G. Travish, R.L. Byer, Demonstration of electron

430 acceleration in a laser-driven dielectric microstructure., Nature. 503 (2013) 91–94.

431 doi:10.1038/nature12664.

432 [5] J. Hsieh, Computed Tomography: Principles, Design, Artifacts, and Recent Advances, SPIE, Bellingham,

433 Washington, USA, 2015.

434 [6] W.A. Kalender, Computed Tomography: Fundamentals, System Technology, Image Quality,

435 Applications, 3rd ed., John Wiley & Sons, Hoboken, New Jersey, USA, 2011.

436 http://eu.wiley.com/WileyCDA/WileyTitle/productCd-389578317X.html.

437 [7] N. Pelc., Recent and future directions in CT Imaging, Ann. Biomed. Eng. 42 (2014) 260–68.

438 [8] E. Shefer, A. Altman, R. Behling, R. Goshen, L. Gregorian, Y. Roterman, I. Uman, N. Wainer, Y. Yagil,

439 O. Zarchin, State of the Art of CT Detectors and Sources: A Literature Review, Curr. Radiol. Rep. 1

440 (2013) 76–91. doi:10.1007/s40134-012-0006-4.

441 [9] E. Eggl, M. Dierolf, K. Achterhold, C. Jud, B. Günther, E. Braig, B. Gleich, F. Pfeiffer, The Munich

442 compact light source: initial performance measures, J Synchrotron Radiat. 23 (2016) 1137–42.

443 doi:10.1107/S160057751600967X.

444 [10] R.E. Alvarez, A. Macovski, Energy-selective reconstructions in X-ray computerized tomography., Phys.

445 Med. Biol. 21 (1976) 733–44. http://www.ncbi.nlm.nih.gov/pubmed/967922 (accessed March 11, 2017).

446 [11] I.P. Almeida, L.E.J.R. Schyns, M.C. Öllers, W. van Elmpt, K. Parodi, G. Landry, F. Verhaegen, Dual-

447 energy CT quantitative imaging: a comparison study between twin-beam and dual-source CT scanners.,

448 Med. Phys. 44 (2017) 171–179. doi:10.1002/mp.12000.

449 [12] A. Pokryvailo, C. Carp, C. Scapellati, A 100 kW high voltage power supply for dual energy computer

450 tomography applications, Proc. 2014 IEEE Int. Power Modul. High Volt. Conf. IPMHVC 2014. (2015)

451 115–118. doi:10.1109/IPMHVC.2014.7287221.

452 [13] R. Raupach, Dual Energy Imaging with Dual Source CT Systems, Med. Phys. 423 (2004) 419–423.
453 [14] S. Kappler, A. Henning, B. Kreisler, F. Schoeck, K. Stierstorfer, T. Flohr, Photon counting CT at elevated

454 X-ray tube currents: contrast stability, image noise and multi-energy performance, Proc. SPIE Med.

455 Imaging. 9033 (2014) 90331C. doi:10.1117/12.2043511.

456 [15] J.C. Santos, A. Tomal, T.A. Furquim, A.M.F. Fausto, M.S. Nogueira, P.R. Costa, Direct measurement of

457 clinical mammographic x-ray spectra using a CdTe spectrometer, Med. Phys. Volume 44 (2017).

458 doi:10.1002/mp.12287.

459 [16] U. Barrett, Harrison H (Department of Radiology and Optical Sciences Center, University of Arizona,

460 Tucson, Az, U. Swindell, William (Department of Radiology and Optical Sciences Center, University of

461 Arizona, Tucson, Az, Radiological Imaging - The theory of image formation, detection, and processing,

462 Academic Press, New York, NY, USA, New York, NY, USA, 1981.

463 [17] R. Raupach, T. Flohr, Performance evaluation of x-ray differential phase contrast computed tomography

464 (PCT) with respect to medical imaging, Med. Phys. 39 (2012) 4761. doi:10.1118/1.4736529.

465 [18] M. Marschner, L. Birnbacher, M. Willner, M. Chabior, A. Fehringer, J. Herzen, P.B. Noël, F. Pfeiffer,

466 Redefining the lower statistical limit in x-ray phase-contrast imaging, Proc. SPIE Med. Imaging. 9412

467 (2015) 94120M. doi:10.1117/12.2082207.

468 [19] T. Weber, G. Pelzer, J. Rieger, A. Ritter, G. Anton, Report of improved performance in Talbot-Lau phase

469 contrast computed tomography, Med. Phys. Lett. 42 (2015) 2892–2896.

470 [20] T. hler, . rgen ngel, E. Roessl, Noise properties of grating-based x-ray phase contrast computed

471 tomography, Med. Phys. 38 (2011) S106. doi:10.1118/1.3532396.

472 [21] K.J. Engel, D. Geller, T. K??hler, G. Martens, S. Schusser, G. Vogtmeier, E. R??ssl, Contrast-to-noise in

473 X-ray differential phase contrast imaging, Nucl. Instruments Methods Phys. Res. Sect. A Accel.

474 Spectrometers, Detect. Assoc. Equip. 648 (2011) 202–207. doi:10.1016/j.nima.2010.11.169.

475 [22] F. Pfeiffer, T. Weitkamp, O. Bunk, C. David, Phase retrieval and differential phase-contrast imaging with

476 low-brilliance X-ray sources, Nat. Phys. 2 (2006) 258–261. doi:10.1038/nphys265.

477 [23] E. Roessl, H. Daerr, T. Koehler, G. Martens, U. van Stevendaal, Clinical boundary conditions for grating-

478 based differential phase-contrast mammography, Philos. Trans. R. Soc. A Math. Phys. Eng. Sci. 372

479 (2014) 1–7. doi:10.1098/rsta.2013.0033.

480 [24] G. Harding, B. Jordan, J. Kosanetzky, A new fluorescent x-ray source for photon scattering

481 investigations, Phys. Med. Biol. 36 (1991) 1573–1583. doi:Doi 10.1088/0031-9155/36/12/003.

482 [25] Nielsen, Jens (2011). Elements of modern X-ray physics. Chichester, West Sussex: John Wiley. ISBN

483 9781119970156.

484 [26] F. Grüner et al., Crossing Sensitivity Frontiers of medical X-ray fluorescence imaging, to be submitted

485 [27] Bernhard H Müller, Christoph Hoeschen, Florian Grüner, Vladimir A Arkadiev, and Thorsten R C

486 Johnson, Molecular imaging based on x-ray fluorescent high-Z tracers, Phys. Med. Biol. 58 (2013) 8063–

487 8076.

488 [28] Donald P. Umstadter, All-laser-driven Thomson X-ray sources, Contemporary Physics 56:4 (2015), pp

489 417–431. doi 10.1080/00107514.2015.1023519.

490 [29] T. Tajima and J.M. Dawson, Laser electron accelerator, Phys. Rev. Lett. 43 (1979), pp. 267–270.

491 [30] C.G.R. Geddes, C. Toth, J. van Tilborg, E. Esarey, C.B. Schroeder, D. Bruhwiler, C. Nieter, J. Cary, and

492 W.P. Leemans, High-quality electron beams from a laser wakefield accelerator using plasma-channel

493 guiding, Nature 431 (2004), pp. 538–541; J. Faure, Y. Glinec, A. Pukhov, S. Kiselev, S. Gordienko, E.

494 Lefebvre, J.-P. Rousseau, F. Burgy, and V. Malka, A laser–plasma accelerator producing monoenergetic

495 electron beams, Nature 431 (2004), pp. 541–544; S.P.D. Mangles, C.D. Murphy, Z. Najmudin, A.G.R.

496 Thomas, J.L. Collier, A.E. Dangor, E.J. Divall, P.S. Foster, J.G. Gallacher, C.J. Hooker, D.A.

497 Jaroszynski, A.J. Langley, W.B. Mori, P.A. Norreys, F.S. Tsung, R. Viskup, B.R. Walton, and K.

498 Krushelnick, Monoenergetic beams of relativistic electrons from intense laser–plasma interactions,

499 Nature 431 (2004), pp. 535–538.

500 [31] V. B. Krapchev, Kinetic Theory of the Ponderomotive Effects in a Plasma, Phys. Rev. Lett. 42, 497

501 (1979)

502 [32] J. M. Cole, J. C. Wood, N. C. Lopes, K. Poder, R. L. Abel, S. Alatabi, J. S. J. Bryant, A. Jin, S. Kneip, K.

503 Mecseki, D. R. Symes, S. P. D. Mangles & Z. Najmudin, Laser-wakefield accelerators as hard x-ray

504 sources for 3D medical imaging of human bone, Scientific Reports 5, Article number: 13244 (2015).

505 [33] J. Faure, C. Rechatin, O. Lundh, L. Ammoura, and V. Malka, Injection and acceleration of

506 quasimonoenergetic relativistic electron beams using density gradients at the edges of a plasma channel,
507 Phys. Plasmas 17 (2010), pp. 1–8, 083107.

508 [34] T. Nubbemeyer, M. Kaumanns, M. Ueffing, M. Gorjan, A. Alismail, H. Fattahi, J. Brons, O. Pronin, H.G.

509 Barros, Z. Major, T. Metzger, D. Sutter, F. rausz, 1  kW, 200  m picosecond thin-disk laser system,

510 Opt. Lett. 42 (2017) 1381. doi:10.1364/OL.42.001381.

511 [35] R. Behling, Performance and Pitfalls of Diagnostic X-Ray Sources : an Overview, Med. Phys. Int. 4

512 (2016) 107–114.