Direct-conversion CdTe/CZT detectors that count and energy-sort every single X-ray photon
For nearly four decades, every clinical CT scanner has used an "indirect" energy-integrating detector (EID): a scintillator crystal converts X-ray photons into visible light, which a photodiode then converts into an analog electrical current. This two-step conversion is robust and inexpensive — but it destroys per-photon energy information and imposes hard limits on spatial resolution and low-dose performance.
A conventional CT detector element performs energy conversion in two discrete physical steps:
• Step 1 — Scintillation: an incident X-ray photon deposits its energy in a scintillator crystal (typically gadolinium oxysulfide, Gd₂O₂S:Pr — "GOS," sometimes doped with cerium and fluorine for faster afterglow decay). The photon's energy is converted into a burst of hundreds to thousands of visible-light photons proportional (on average) to the absorbed energy. • Step 2 — Photodetection: the scintillation light spreads isotropically through the crystal and reflective septa before striking a silicon photodiode bonded beneath the scintillator. The photodiode integrates the light into an analog current over the sampling interval, which is amplified and digitized.
Because the light spreads before collection, each detector element unavoidably picks up some light generated by photons absorbed in neighboring crystal elements, softening the effective point-spread function of the array — this optical crosstalk is a primary reason conventional pixel pitch cannot easily shrink below ~0.5–1.0 mm without severe efficiency and crosstalk penalties.
An energy-integrating detector sums the total light output of every photon that arrives during the sampling window into a single number. A 40 keV photon and a 120 keV photon contribute charge in proportion to their energy, but the detector electronics have no way to separately tag or bin them — the output is one aggregate signal per pixel per projection.
This creates two well-known artifacts:
• Beam hardening: as a polychromatic X-ray beam passes through tissue, lower-energy photons are preferentially absorbed, shifting the mean beam energy higher ('hardening' it) — energy-integrating systems cannot correct this per-photon, only through empirical calibration and iterative correction algorithms. • Energy-weighting bias: energy-integrating detection over-weights high-energy photons (which deposit more light per event) relative to their actual contribution to image contrast, worsening the achievable contrast-to-noise ratio (CNR) per unit dose compared to a detector that could count and appropriately weight every photon individually.
To obtain any spectral information at all, conventional CT historically required dual-energy techniques external to the detector — fast kVp switching, dual-source geometry, or dual-layer sandwich detectors — adding hardware complexity, extra dose, or temporal misregistration.
Every analog readout chain — photodiode, transimpedance amplifier, integrator, analog-to-digital converter — adds electronic (thermal, shot, and quantization) noise to the signal regardless of how many photons actually arrived. This noise floor is added to every reading, every projection, every rotation.
At low dose (few photons per detector element per view — the exact regime clinicians want for pediatric, screening, and repeat-scan protocols) the electronic noise floor becomes a larger fraction of the total signal variance, degrading image quality disproportionately. This is precisely the physical weakness that photon-counting detectors are engineered to remove.
Gadolinium oxysulfide (GOS) scintillators have been the CT detector workhorse since the 1990s, but their inherent optical light spread caps practical pixel pitch near 0.5–1.0 mm and their two-step conversion chain caps spectral information at a single, energy-blind channel.
Photon-counting detectors (PCDs) discard the scintillator entirely. A wafer of cadmium telluride (CdTe) or cadmium zinc telluride (CZT) is biased with a strong electric field across its thickness; an absorbed X-ray photon creates a compact cloud of electron-hole pairs directly, with no optical step, no light spreading, and a charge magnitude that is intrinsically proportional to the photon's own energy.
When an X-ray photon is absorbed in CdTe or CZT — predominantly via the photoelectric effect at diagnostic CT energies — it ejects a photoelectron that rapidly loses energy through ionization, creating thousands of electron-hole pairs in a compact cloud roughly tens of micrometers across. A strong applied bias field (typically 500–1000 V across a ~1.6 mm thick crystal) sweeps electrons toward the pixelated anode and holes toward the cathode in a few nanoseconds.
The number of electron-hole pairs generated is directly proportional to the absorbed photon energy (roughly one pair per ~4.4 eV in CdTe), so the amplitude of the resulting current pulse at the pixel electrode is itself an analog measurement of that single photon's energy — a fundamentally different information channel than the light-integration approach of scintillators.
Because the conversion happens in one solid-state step with no light diffusion, the pulse is collected almost entirely under the pixel where the photon landed, with only small, well-characterized charge-sharing effects at pixel boundaries — enabling the dramatically smaller pixel pitch (150–225 µm) that underlies PCD spatial resolution gains.
CdTe and CZT (a zinc-alloyed variant, Cd₁₋ₓZnₓTe) are chosen for CT detectors because they combine several properties simultaneously required for clinical X-ray energies (roughly 20–140 keV):
• High effective atomic number (Z_eff ≈ 50) gives strong photoelectric absorption efficiency across the diagnostic range, unlike silicon (Z=14), which is essentially transparent to CT-energy X-rays • High density (~5.8–6.2 g/cm³) further boosts attenuation per unit thickness, allowing a thin (~1.5–2 mm) crystal to stop the vast majority of incident photons • Wide bandgap (CdTe ≈ 1.44 eV, CZT ≈ 1.6 eV) keeps thermally generated leakage (dark) current low enough for room-temperature operation, unlike narrower-gap semiconductors • Favorable charge transport (electron mobility ~1000 cm²/V·s in CdTe) allows fast charge collection, which is essential for handling clinical photon flux rates without excessive charge trapping or polarization
CZT offers somewhat lower leakage current and better long-term stability against polarization (a drift phenomenon from trapped charge that can degrade CdTe performance over hours of continuous high-flux operation), at a modest cost in absorption efficiency versus pure CdTe — vendors select between the two based on these trade-offs.
The K-edges of cadmium (26.7 keV) and tellurium (31.8 keV) sit directly within the diagnostic energy window, producing an absorption efficiency discontinuity that detector engineers exploit when placing the lowest energy-bin thresholds for maximal iodine and soft-tissue contrast sensitivity.
Underneath every pixel, dedicated application-specific integrated circuit (ASIC) electronics perform, in real time and for every single detected photon:
1. Charge-sensitive preamplification of the tiny (femtocoulomb-scale) current pulse 2. Pulse shaping to a defined width (tens of nanoseconds) matched to expected clinical count rates 3. Comparison of the peak pulse height against multiple voltage thresholds simultaneously 4. Incrementing the appropriate energy-bin counter for that pixel
This entire chain — happening independently in every one of the hundreds of thousands of pixels on a detector module, at up to tens of millions of photon events per mm² per second — is what allows a photon-counting CT detector to report not just how much energy arrived, but literally how many photons arrived and in which energy ranges, every rotation, every projection.
| Product | Indication | Trial Design | Key Result |
|---|---|---|---|
| GOS / GdOS Scintillator | Conventional EID CT | Photon → visible light → photodiode current; light spreads before collection | Mature, inexpensive, robust — the 40-year clinical standard |
| CdTe (Cadmium Telluride) | PCD-CT (e.g. NAEOTOM Alpha) | Photon → e⁻/h⁺ cloud directly under bias field; ~4.4 eV per pair | High Z_eff (50), strong absorption, fast collection |
| CZT (Cd₁₋ₓZnₓTe) | PCD-CT (alternate material) | Zinc-alloyed CdTe; wider bandgap (~1.6 eV), lower leakage current | Better polarization stability at high flux, lower dark current |
| Silicon PCD (emerging) | Research / photon-counting mammography | Low Z (14); requires edge-on or very thick geometry to absorb CT photons | Excellent charge transport, very low noise, mature fabrication |
The defining capability of photon-counting CT is that every individual absorbed photon is sorted into one of typically 2–5 energy bins the instant it is detected, using a bank of fast voltage comparators referenced to programmable thresholds. This is fundamentally different from post-hoc energy weighting — the sorting happens at the pixel, in real time, before any spatial or temporal averaging occurs.
Each pixel's ASIC channel contains a small bank of analog comparators, each referenced to an independently programmable voltage threshold corresponding to a chosen photon energy (since pulse height is proportional to deposited energy). When a shaped charge pulse arrives:
• The pulse height is compared simultaneously against every threshold • The highest threshold the pulse exceeds determines its energy bin • A digital counter for that specific bin, in that specific pixel, is incremented by one • The process resets in tens of nanoseconds, ready for the next photon
A typical clinical configuration might set thresholds at roughly 20 keV (the low-energy noise/scatter rejection cut), 45 keV, 65 keV, and 90 keV, producing four or five effective bins covering the diagnostic 20–140 keV window. Ultra-high-resolution acquisition modes on commercial scanners sometimes trade bin count for smaller effective pixel size, using threshold ladders such as 20/25/45/55/65/70/90 keV depending on vendor and clinical protocol.
Threshold placement is not arbitrary — it is chosen to maximize diagnostic value:
• A low threshold (~20 keV) rejects electronic noise and very-low-energy scatter without discarding useful photons • A threshold near the iodine K-edge (33.2 keV) sharply improves iodine contrast-to-noise, since iodine's attenuation coefficient jumps discontinuously there — critical for CT angiography and contrast-enhanced oncologic imaging • Thresholds bracketing the Cd (26.7 keV) and Te (31.8 keV) K-edges of the detector material itself must be managed carefully, since K-escape and fluorescence events near these energies can shift counts between neighboring pixels and bins — a well-characterized correction is built into reconstruction • A high-energy bin (>90 keV) isolates the most tissue-penetrating photons, useful for calcium/iodine discrimination and reducing beam-hardening artifacts from dense bone or stents
Because every acquisition already carries 2–5 independent energy-weighted projection datasets, material decomposition (soft tissue / iodine / calcium) is available from a single scan — without dual-source hardware or kVp switching.
Comparator-based counting requires each photon's pulse to be temporally resolved from its neighbors. At very high photon flux (thick patients at low kVp, or unattenuated beam near the source), two photons can arrive so close together that their pulses overlap ("pile-up"), and the electronics may either miscount them as one higher-energy photon (energy pile-up, corrupting the spectrum) or fail to register the second pulse at all (count-rate losses).
Modern PCD-CT ASICs mitigate this with sub-50 ns dead times per pixel and, in the case of NAEOTOM Alpha-class systems, small individual pixel areas so that the incident flux per pixel — rather than per detector module — stays within the linear counting regime even at typical clinical mA settings. Very high flux still requires either automatic exposure control adjustments or acceptance of a graceful, corrected loss of count-rate linearity at the highest tube currents.
Because every photon is discretely counted rather than integrated, photon-counting detectors also achieve intrinsically higher detective quantum efficiency (DQE) at low dose — every registered count carries a full, weighted contribution to the image, rather than being diluted by an electronic noise floor.
Two of the most clinically consequential advantages of photon-counting CT flow from the same underlying architecture change: because sub-threshold pulses are simply discarded rather than digitized, there is no electronic noise floor added to the signal; and because charge clouds stay compact under the pixel where they were generated, pixel pitch can shrink to 150–225 µm — roughly one-quarter to one-third the size of conventional detector elements.
In an energy-integrating detector, electronic noise (thermal noise in the amplifier, dark current in the photodiode, quantization noise in the ADC) is summed into the analog signal continuously — it cannot be separated from real photon signal after the fact, and it is present in every single readout, dose or no dose.
In a photon-counting detector, the lowest comparator threshold acts as a hard discriminator: a pulse must exceed a chosen voltage (energy) to register as a count at all. Small-amplitude electronic noise fluctuations that never cross this threshold are never counted, never digitized, never summed into the image. The counted signal is, by construction, built almost entirely from real photon events above the noise threshold.
The clinical consequence is disproportionate: at low photon flux (i.e., low-dose acquisitions), where the noise floor would otherwise represent a comparatively large fraction of total signal variance in an EID system, the PCD system's noise remains dominated by genuine quantum (Poisson) statistics of the X-ray beam itself — the theoretical noise floor of any imaging system — rather than by added electronics.
Clinical and phantom studies comparing photon-counting CT to conventional energy-integrating CT at matched image quality have reported:
• Dose reductions of roughly 20–45% at equivalent contrast-to-noise ratio, with the largest gains in low-dose and pediatric protocols where the EID electronic noise floor penalty is largest • Improved iodine contrast-to-noise ratio per unit dose, from both optimal energy weighting (low-energy photons — more heavily attenuated by iodine — can be preferentially weighted rather than diluted by high-energy photons) and the absence of electronic noise • Reduced image noise texture at a given dose level, improving low-contrast detectability for subtle lesions
Because these gains stack with the ability to reconstruct at smaller pixel sizes without a corresponding noise penalty (since there is no scintillator light-spread crosstalk to fight), photon-counting CT can simultaneously offer lower dose and higher spatial resolution — a combination that was a fundamental trade-off on conventional EID systems.
Conventional EID detector elements are typically 0.5–1.0 mm at isocenter, constrained largely by scintillator light spread and photodiode manufacturing/crosstalk limits. Direct-conversion PCD pixels avoid the optical step entirely, so charge clouds remain tightly localized (tens of micrometers), allowing physical pixel pitches of 150–225 µm.
Commercial systems exploit this in two ways:
• Standard resolution mode: multiple adjacent small physical pixels are summed on-chip into a larger effective macro-pixel, matching conventional detector geometry while retaining the noise and spectral advantages • Ultra-High-Resolution (UHR) mode: the small physical pixels are read out individually (or in smaller groups), yielding reconstructed in-plane spatial resolution as fine as 0.11–0.2 mm — roughly 2–3× finer than conventional CT — at some cost in per-pixel photon statistics that is offset by the elimination of electronic noise.
UHR mode is particularly valuable for high-contrast, fine-structure imaging: temporal bone / inner-ear anatomy, lung parenchyma (bronchiolar walls, fine septal detail), and coronary stent lumens, where conventional detector blur is the limiting factor rather than dose.
Because pixel pitch and noise floor are decoupled from dose in a way that was impossible with scintillator detectors, photon-counting CT is often described as breaking the historical trade-off triangle between spatial resolution, image noise, and radiation dose.
Because every photon-counting CT acquisition already sorts photons into multiple energy bins, spectral results — virtual monoenergetic images, material-specific density maps, and calcium suppression — are available from every single scan, with no dual-source hardware, no kVp switching, and no extra dose. The Siemens NAEOTOM Alpha, FDA-cleared in September 2021, was the first clinical-routine photon-counting CT scanner, and has since driven adoption in coronary and pulmonary imaging.
Conventional dual-energy CT requires one of several hardware or acquisition workarounds to obtain two energy datasets: rapid kVp switching (alternating tube voltage between projections, at the cost of temporal registration and flux limitations), dual-source geometry (two X-ray tube/detector pairs at different kVp, at the cost of a second full detector system), or dual-layer "sandwich" detectors (a low-energy-absorbing front layer atop a high-energy-absorbing back layer, at the cost of some spectral separation quality).
Photon-counting CT sidesteps all of these: because each pixel already reports counts in 2–5 distinct energy bins from a single exposure at a single kVp, the projection data itself is inherently multi-energy. A single, standard acquisition yields everything needed for spectral post-processing — no protocol change, no extra dose, no registration error between energy datasets acquired at different times.
From the multi-bin projection data, reconstruction software derives several clinically used spectral products:
• Virtual monoenergetic images (VMI): synthetic images representing what the scan would look like at a single, user-selected photon energy (e.g., 40–190 keV), computed by appropriately weighting the multiple energy-bin datasets. Low-keV VMI (~40–55 keV) boosts iodine contrast; high-keV VMI (~130–190 keV) reduces beam-hardening and metal artifacts. • Material decomposition maps: using 2–3 basis materials (typically water, iodine, and sometimes calcium or hydroxyapatite), each voxel's attenuation is decomposed into material-specific density maps — an iodine map directly visualizes contrast distribution (e.g., myocardial perfusion, pulmonary emboli) independent of the underlying soft tissue. • Calcium/iodine separation and calcium suppression: because calcium and iodine have distinguishable spectral signatures, iodine enhancement can be visualized with calcified plaque or bone digitally suppressed or separated — valuable in coronary CT angiography, where dense calcium blooming can obscure the true arterial lumen. • Effective atomic number (Z_eff) maps: useful for characterizing renal stone composition or differentiating tissue types beyond simple density.
One of the most clinically impactful photon-counting CT applications is coronary CT angiography. On conventional CT, densely calcified plaque and metallic coronary stents produce "blooming" artifact — the apparent size of the high-attenuation structure is overestimated due to detector blur, partial volume averaging, and beam hardening, often obscuring the true lumen diameter and causing false-positive stenosis calls.
Photon-counting CT reduces blooming through the combination of smaller pixel pitch (sharper point-spread function, less partial-volume blur), optimal energy weighting (better differentiation between calcium/metal and iodine-enhanced lumen), and high-keV virtual monoenergetic reconstruction (reduced beam-hardening streaks from calcium and stent struts). Published comparisons report visibly improved in-stent lumen visualization and more accurate stenosis grading in heavily calcified vessels compared to energy-integrating CT.
The Siemens NAEOTOM Alpha, the first FDA-cleared photon-counting CT scanner for clinical routine use (cleared September 2021), pairs two CdTe-based photon-counting detector/source systems in a dual-source gantry. It offers standard resolution acquisition alongside an Ultra-High-Resolution (UHR) mode using the full small physical pixel pitch, reaching a maximum reconstructed isotropic resolution around 0.66 mm (with in-plane detail resolvable to roughly 0.15–0.2 mm depending on reconstruction kernel and target).
In thoracic/lung imaging, this resolution gain directly benefits visualization of fine anatomic structures: bronchiolar walls, interlobular septa, small airways disease, and subtle ground-glass or fibrotic patterns relevant to interstitial lung disease — structures historically at or below the resolving power of conventional 0.5–0.6 mm CT detectors. Combined with the dose-reduction properties of noise-floor-free counting, UHR lung protocols can pursue finer anatomic detail without a proportional dose penalty, an especially valuable trade for patients requiring serial follow-up imaging.
Since the NAEOTOM Alpha's 2021 FDA clearance, photon-counting CT has moved from a research concept — first demonstrated in prototype scanners in the 2000s–2010s — into routine clinical service across cardiology, pulmonology, and oncologic imaging, with additional vendors developing competing photon-counting platforms.