Spectral imaging (radiography)
Spectral imaging is an umbrella term for energy-resolved
Background
The first medical application of spectral imaging appeared in 1953 when B. Jacobson at the
In recent years, however, two fields of technological breakthrough have spurred a renewed interest in energy-resolved imaging. Firstly, single-scan energy-resolved CT was introduced for routine clinical use in 2006 and is now available by several major manufacturers,
Spectral image acquisition
An energy-resolved imaging system probes the object at two or more photon energy levels. In a generic imaging system, the projected signal in a detector element at energy level is[1]
-
(1)
where is the number of incident photons, is the normalized incident energy spectrum, and is the detector response function.
Most elements appearing naturally in human bodies are of low atomic number and lack absorption edges in the diagnostic X-ray energy range. The two dominating X-ray interaction effects are then Compton scattering and the photo-electric effect, which can be assumed to be smooth and with separable and independent material and energy dependences. The linear attenuation coefficients can hence be expanded as[6]
-
(2)
In contrast-enhanced imaging, high-atomic-number contrast agents with K absorption edges in the diagnostic energy range may be present in the body. K-edge energies are material specific, which means that the energy dependence of the photo-electric effect is no longer separable from the material properties, and an additional term can be added to Eq. (2) according to[12]
-
(3)
where and are the material coefficient and energy dependency of contrast-agent material .
Energy weighting
Summing the energy bins in Eq. (1) () yields a conventional non-energy-resolved image, but because X-ray contrast varies with energy, a weighted sum () optimizes the contrast-to-noise-ratio (CNR) and enables a higher CNR at a constant patient dose or a lower dose at a constant CNR.[13] The benefit of energy weighting is highest where the photo-electric effect dominates and lower in high-energy regions dominated by Compton scattering (with weaker energy dependence).
Energy weighting was pioneered by Tapiovaara and Wagner[13] and has subsequently been refined for projection imaging[14][15] and CT[16] with CNR improvements ranging from a few percent up to tenth of percent for heavier elements and an ideal CT detector.[17] An example with a realistic detector was presented by Berglund et al. who modified a photon-counting mammography system and raised the CNR of clinical images by 2.2–5.2%.[18]
Material decomposition
Equation (1) can be treated as a system of equations with material thicknesses as unknowns, a technique broadly referred to as material decomposition. System properties and linear attenuation coefficients need to be known, either explicitly (by modelling) or implicitly (by calibration). In CT, implementing material decomposition post reconstruction (image-based decomposition) does not require coinciding projection data, but the decomposed images may suffer from beam-hardening artefacts because the reconstruction algorithm is generally non-reversible.[19] Applying material decomposition directly in projection space instead (projection-based decomposition),[6] can in principle eliminate beam-hardening artefacts because the decomposed projections are quantitative, but the technique requires coinciding projection data such as from a detection-based method.
In the absence of K-edge contrast agents and any other information about the object (e.g. thickness), the limited number of independent energy dependences according to Eq. (2) means that the system of equations can only be solved for two unknowns, and measurements at two energies () are necessary and sufficient for a unique solution of and .[7] Materials 1 and 2 are referred to as basis materials and are assumed to make up the object; any other material present in the object will be represented by a linear combination of the two basis materials.
Material-decomposed images can be used to differentiate between healthy and malignant tissue, such as
The basis-material representation can be readily converted to images showing the amounts of photoelectric and Compton interactions by invoking Eq. (2), and to images of effective-atomic-number and electron density distributions.[6] As the basis-material representation is sufficient to describe the linear attenuation of the object, it is possible to calculate virtual monochromatic images, which is useful for optimizing the CNR to a certain imaging task, analogous to energy weighting. For instance, the CNR between grey and white brain matter is maximized at medium energies, whereas artefacts caused by photon starvation are minimized at higher virtual energies.[31]
K-edge imaging
In
Enhancement of contrast agents can be used for improved detection and diagnosis of tumors,
Most studies of contrast-enhanced spectral imaging have used iodine, which is a well-established contrast agent, but the K edge of iodine at 33.2 keV is not optimal for all applications and some patients are hypersensitive to iodine. Other contrast agents have therefore been proposed, such as gadolinium (K edge at 50.2 keV),[39] nanoparticle silver (K edge at 25.5 keV),[40] zirconium (K edge at 18.0 keV),[41] and gold (K edge at 80.7 keV).[42] Some contrast agents can be targeted,[43] which opens up possibilities for molecular imaging, and using several contrast agents with different K-edge energies in combination with photon-counting detectors with a corresponding number of energy thresholds enable multi-agent imaging.[44]
Technologies and methods
Incidence-based methods obtain spectral information by acquiring several images at different
Detection-based methods instead obtain spectral information by splitting the spectrum after interaction in the object. So-called sandwich detectors consist of two (or more) detector layers, where the top layer preferentially detects low-energy photons and the bottom layer detects a harder spectrum.[47][48] Detection-based methods enable projection-based material decomposition because the two energy levels measured by the detector represent identical ray paths. Further, spectral information is available from every scan, which has work-flow advantages.[49]
The currently most advanced detection-based method is based on
The MicroDose system was based on silicon strip detectors, are currently quite far from clinical implementation.
The main intrinsic challenge of photon-counting detectors for medical imaging is pulse pileup,[62] which results in lost counts and reduced energy resolution because several pulses are counted as one. Pileup will always be present in photon-counting detectors because of the Poisson distribution of incident photons, but detector speeds are now so high that acceptable pileup levels at CT count rates begin to come within reach.[67]
See also
References
- ^ S2CID 125589580.
- S2CID 5871406.
- S2CID 31601055.
- PMID 13079943.
- S2CID 5820281.
- ^ S2CID 250824716.
- ^ PMID 7290019.
- PMID 15070254.
- ^ S2CID 628323.
- ^ PMID 17654894.
- PMID 26840654.
- ^ Šuković, P.; Clinthorne, N.H. (1999). "Basis material decomposition using triple-energy Xray computed tomography". Proceedings of the 16th IEEE Instrumentation and Measurement Technology Conference, IMTC/99, Venice, Italy: 1615–1618.
- ^ S2CID 250758224.
- PMID 10619253.
- .
- S2CID 7791460.
- S2CID 17685742.
- PMID 26158045.
- PMID 19746815.
- S2CID 38228845.
- PMID 15984691.
- S2CID 53717425.
- PMID 20551186.
- PMID 20807846.
- S2CID 24872066.
- PMID 28421611.
- PMID 11997567.
- PMID 20229889.
- PMID 26001233.
- PMID 18599239.
- S2CID 3881271.
- PMID 12888621.
- S2CID 25228858.
- PMID 24874766.
- PMID 20679449.
- PMID 19487466.
- PMID 25070169.
- PMID 27863922.
- PMID 28124106.
- ^ Karunamuni, R.; Al Zaki, A.; Popov, A.V.; Delikatny, E.J.; Gavenonis, S.; Tsourkas, A.; Maidment, A.D.A. (2012). "An examination of silver as a radiographic contrast agent in dualenergy breast X-ray imaging, IWDM 2012, LNCS". 7361: 418–425.
{{cite journal}}
: Cite journal requires|journal=
(help) - S2CID 28937454.
- PMID 23185109.
- PMID 20668118.
- PMID 27918709.
- PMID 21520830.
- .
- PMID 12468068.
- .
- S2CID 20185571.
- PMID 24089889.
- ^ S2CID 121348971.
- )
- S2CID 22734564.
- PMID 23127077.
- .
- S2CID 34780307.
- S2CID 121200701.
- .
- PMID 19920884.
- S2CID 34260079.
- PMID 19994521.
- ^ a b Knoll, G.F. (2000). Radiation Detection and Measurement. John Wiley & Sons.
- S2CID 43453671.
- PMID 19472640.
- PMID 20161098.
- S2CID 72756191.
- PMID 26835839.