2. CT Physics

Every CT number begins as kinetic energy delivered to electrons inside an evacuated rotating-anode tube. A heated tungsten or tungsten–rhenium filament liberates electrons by thermionic emission, and a potential difference of typically $80$–$140\,\mathrm{kV}$ accelerates them across the tube toward the anode. The kinetic energy each electron acquires is set entirely by the tube potential, $E_k = eV$, so an electron crossing a $120\,\mathrm{kV}$ gap arrives with $120\,\mathrm{keV}$. The tube current (quantified clinically as the time integral, the mAs) fixes the number of electrons and therefore the quantity of photons, whereas the kilovoltage fixes their maximum and mean energy. This decoupling—mAs governs flux and quantum noise, kVp governs spectral hardness and subject contrast—is the lever every CT protocol pulls, and misunderstanding it is the root of many dose–quality errors.

On striking the anode, fewer than one percent of electrons produce useful x-rays; the remainder deposit heat, which is why anode thermal capacity and rotation, not photon yield, limit sustained output. The dominant radiative process is bremsstrahlung (“braking radiation”): an incident electron decelerates in the Coulomb field of a tungsten nucleus and emits a photon whose energy equals the kinetic energy lost in that single encounter. Because the impact parameter varies continuously, the emitted spectrum is continuous, extending from near zero up to a sharp maximum at the Duane–Hunt limit corresponding to a complete stop, $E_{\max} = eV_{\mathrm{peak}}$. The unfiltered bremsstrahlung fluence falls roughly linearly with photon energy, $I(E)\propto Z(E_{\max}-E)$, but the spectrum that actually leaves the tube is reshaped by inherent and added filtration. Low-energy photons, which would only deposit skin dose without penetrating to the detector, are preferentially absorbed; this beam hardening at the source raises the mean energy to roughly $50$–$70\%$ of the peak and is the deliberate purpose of the bow-tie and flat aluminum/copper filters.

Superimposed on the continuum are discrete characteristic lines. When an incident electron ejects an inner-shell (K-shell) tungsten electron, an outer-shell electron fills the vacancy and the binding-energy difference is released as a photon of fixed energy—$59.3\,\mathrm{keV}$ ($K\alpha_1$) and $67.2\,\mathrm{keV}$ ($K\beta$) for tungsten. These lines appear only when the tube potential exceeds the $69.5\,\mathrm{keV}$ K-edge binding energy; below that threshold the spectrum is purely bremsstrahlung. The characteristic peaks contribute a useful, relatively monochromatic burst of mid-energy photons that improves penetration efficiency.

The expert consequence is that the CT beam is irreducibly polychromatic, and this single fact propagates through the entire chain: it makes attenuation energy-dependent, makes the Hounsfield unit kVp-dependent, and creates beam-hardening artifacts because the spectrum continues to harden as it traverses the patient. Quantitatively, the effective energy of a $120\,\mathrm{kVp}$ beam after patient filtration is often $60$–$75\,\mathrm{keV}$, not $120\,\mathrm{keV}$, and it is this effective energy—not the nominal kVp—that determines tissue contrast. Reading CT numbers without holding the spectral context in mind is the first and most common conceptual failure mode in quantitative interpretation.

Once produced, the polychromatic beam is reshaped not by transmission alone but by two competing absorption mechanisms whose relative weight encodes the diagnostic information of CT. Macroscopically, the transmitted intensity through a homogeneous thickness $x$ obeys the Beer–Lambert law, $I = I_0 e^{-\mu x}$, where $\mu$ is the linear attenuation coefficient. For the heterogeneous patient this generalizes to a path integral, $I = I_0 \exp\!\left(-\int \mu(s)\,ds\right)$, and the projection actually measured is its logarithm, $p = -\ln(I/I_0) = \int \mu(s)\,ds$. Reconstruction inverts this line-integral set to recover the spatial map $\mu(\mathbf{r})$; the entire image is therefore a map of attenuation coefficients, and everything clinical follows from how $\mu$ depends on tissue composition and photon energy.

Total attenuation in the diagnostic range ($\sim 20$–$150\,\mathrm{keV}$) is the sum of two dominant terms, $\mu = \mu_{\mathrm{PE}} + \mu_{\mathrm{Compton}}$ (coherent scatter and, at these energies, pair production being negligible). The photoelectric effect dominates at lower energies and in higher-$Z$ materials: an incoming photon is wholly absorbed by ejecting a bound inner-shell electron, possible only when $E_\gamma$ exceeds that shell's binding energy. Its cross-section scales steeply, approximately $\mu_{\mathrm{PE}} \propto \rho\, Z^{3\text{--}4}/E^3$. The cubic inverse-energy dependence explains why iodine ($Z=53$) and barium ($Z=56$) generate enormous contrast at low keV, and the $Z^{3\text{--}4}$ dependence explains why cortical bone and calcium are so conspicuous. Crucially, the cross-section is not smooth: at the binding energy of each shell it jumps discontinuously—the K-edge. Iodine's K-edge at $33.2\,\mathrm{keV}$ and barium's at $37.4\,\mathrm{keV}$ are the physical foundation of spectral material discrimination, because just above the edge the photoelectric absorption of that element rises abruptly while neighboring tissues do not.

Compton scattering dominates at the higher diagnostic energies and in soft tissue. Here the photon interacts with an essentially free outer-shell electron, transferring part of its energy and scattering through angle $\theta$ with the Compton wavelength shift $\Delta\lambda = \frac{h}{m_e c}(1-\cos\theta)$. The Compton cross-section depends on electron density, hence on physical density, but is nearly independent of atomic number and only weakly dependent on energy. This near-$Z$-independence is why soft tissues with very different elemental makeup but similar density (muscle, solid organs, blood) crowd together near $30$–$60\,\mathrm{HU}$ on a conventional scan, and why intravenous iodinated contrast—adding a photoelectric absorber—is so often necessary to separate them. Compton-scattered photons that reach the detector degrade contrast and are the physical origin of scatter artifact, countered by anti-scatter grids and post-hoc correction.

The expert reasoning is to think in two channels at once. Where the question is composition or elemental identity (calcium versus iodine, fat versus blood), photoelectric/$Z$-dependent behavior at low energy carries the signal—and dual-energy imaging exploits exactly this. Where the question is bulk density (consolidation versus aeration, fluid versus solid), Compton behavior dominates and is nearly energy-flat. The polychromatic beam means $\mu$ is an energy-weighted average over the spectrum, so as the beam hardens along a ray the effective $\mu$ of a fixed tissue falls, producing cupping and streak beam-hardening artifacts that the reader must distinguish from true low-attenuation pathology.

The reconstructed map of linear attenuation coefficients is scanner-, energy-, and kVp-specific and therefore not directly transferable; Hounsfield's insight was to normalize it. The Hounsfield unit (CT number) rescales each voxel's coefficient against water by a two-point calibration:

$\mathrm{HU} = 1000 \times \frac{\mu_{\mathrm{tissue}} - \mu_{\mathrm{water}}}{\mu_{\mathrm{water}} - \mu_{\mathrm{air}}}.$

By construction water is fixed at $0\,\mathrm{HU}$ and air at $-1000\,\mathrm{HU}$, anchoring a linear scale on which one HU corresponds to a $0.1\%$ change in $\mu$ relative to water. Because the calibration is relative, much—but not all—of the scanner-to-scanner and kVp-to-kVp variation in $\mu$ cancels for tissues near water density. It does not fully cancel for materials whose photoelectric behavior differs from water (iodine, calcium, bone), whose CT numbers consequently rise as kVp falls: iodine that measures $\sim300\,\mathrm{HU}$ at $120\,\mathrm{kVp}$ may exceed $\sim450\,\mathrm{HU}$ at $80\,\mathrm{kVp}$ because the photoelectric term grows steeply at lower effective energy. The disciplined reader therefore treats an HU value as meaningful only alongside its acquisition kVp, reconstruction kernel, and contrast phase.

The characteristic ranges below organize most diagnostic reasoning. They are approximate and reference $120\,\mathrm{kVp}$, soft-tissue kernel.

Tissue / material	Typical HU
Air	$-1000$
Lung parenchyma	$-950$ to $-700$
Fat	$-120$ to $-60$
Water / simple cyst	$-10$ to $+10$
Most soft tissue / muscle	$+30$ to $+60$
Unenhanced acute hematoma	$+50$ to $+80$
Enhancing parenchyma (post-contrast)	$+80$ to $+120$
Calcium / cortical bone	$+300$ to $+2000$
Metal (implant, electrode)	$> +3000$ (often clipped)

The clinical power lies in using narrow HU thresholds as quantitative decision rules, each rooted in the physics above. A homogeneous lesion measuring $< 0\,\mathrm{HU}$ contains macroscopic fat and, in the right organ, all but secures a diagnosis (renal angiomyolipoma, hepatic or adrenal lipid-rich lesion). The adrenal adenoma rule is paradigmatic: lipid-rich cytoplasm lowers unenhanced attenuation to $\le 10\,\mathrm{HU}$ with high specificity; when above threshold, a contrast washout calculation—absolute washout $\frac{\mathrm{HU}_{\mathrm{enh}}-\mathrm{HU}_{\mathrm{delayed}}}{\mathrm{HU}_{\mathrm{enh}}-\mathrm{HU}_{\mathrm{unenh}}}\times100\%$ exceeding $\sim60\%$—reclassifies the lesion as a benign adenoma by its vascular kinetics rather than its lipid. Acute hemorrhage is conspicuous at $50$–$80\,\mathrm{HU}$ because clot retraction concentrates dense protein; as the clot lyses over days it falls toward and below water, so attenuation timestamps the bleed. A simple cyst is held to $\le 20\,\mathrm{HU}$ and must not enhance, the distinction between a hyperdense cyst and an enhancing nodule turning on a $10$–$20\,\mathrm{HU}$ difference that demands matched pre/post technique.

Two failure modes recur. Partial-volume averaging mixes the attenuation of two materials sharing a voxel, so a small calcification beside fat can read as bland soft tissue, or a tiny enhancing focus can be diluted below threshold; thin sections mitigate this. Pseudo-enhancement, a beam-hardening byproduct, artifactually raises the apparent HU of a renal cyst adjacent to densely opacified cortex, fabricating enhancement that is physical, not biological. Recognizing that an HU is an energy-weighted, partial-volume–blurred estimate—not a ground truth—separates quantitative competence from naive number-reading.

After traversing the patient, the surviving photons must be converted to a measurable electrical signal with high fidelity, because detector imperfections add directly to image noise and dose. The dominant clinical architecture is the energy-integrating scintillation detector: a high-$Z$ ceramic scintillator (gadolinium oxysulfide, $\mathrm{Gd_2O_2S}$, or rare-earth garnets such as gadolinium gallium garnet) absorbs an x-ray photon and re-emits its energy as a burst of visible light, which a coupled photodiode converts to current. The signal integrated over a sampling interval is proportional to the total deposited energy, $S \propto \sum_i E_i$, which is the defining limitation: a single $100\,\mathrm{keV}$ photon and two $50\,\mathrm{keV}$ photons are indistinguishable, so all spectral information carried by individual photon energies is discarded, and low-energy photons—though they carry the most tissue contrast—are weighted least.

Detector performance is captured by several efficiencies whose product sets the detective quantum efficiency (DQE), the fraction of incident-photon information actually preserved. Absorption (quantum) efficiency is the probability that an incident photon is stopped in the scintillator, governed by $1 - e^{-\mu_{\mathrm{det}} t}$ and engineered close to unity by adequate scintillator thickness and high $Z$. Geometric efficiency is the fraction of the detector face that is photosensitive rather than occupied by inter-element septa and dead space; reflective septa prevent optical crosstalk and reject scatter but consume area, so the fill factor is a deliberate compromise. Conversion and coupling efficiency describe scintillation light yield and its transfer to the photodiode. Two temporal properties matter clinically: afterglow (residual luminescence) must be minimized to prevent shadowing artifacts at the high gantry rotation speeds of cardiac CT, and the primary speed must support sub-second rotations. Two noise sources compete—quantum (Poisson) noise scaling as $\sqrt{N}$ with photon number, and electronic noise added by the photodiode and readout. At low signal (large patients, low-dose protocols, the shoulders, the posterior fossa), electronic noise can dominate and produces photon-starvation streaks, one of the chief artifacts the reader attributes to detector physics rather than anatomy.

The transition from single-slice to multidetector CT (MDCT) reorganized the detector into a two-dimensional array of many contiguous rows along the patient ($z$) axis. With $64$, $128$, $256$, or up to $320$ rows of sub-millimeter elements, a single rotation now samples a wide $z$-volume simultaneously, enabling isotropic voxels, true multiplanar reconstruction, and whole-organ coverage in one gantry turn. This is the physical enabler of CT angiography and perfusion, where temporal resolution and anatomic coverage must coexist. Wide arrays introduce their own physics problems—the diverging cone beam departs from the idealized fan geometry, demanding cone-beam reconstruction and creating cone-beam artifacts toward the array edges, and scatter rises with the irradiated volume, requiring more aggressive anti-scatter grids and software correction.

The interactive below uses a real, true-HU body CT to make the detector–display chain tangible: switching between mediastinal and lung windows shows how the same recorded attenuation data, captured by the same detector, yields entirely different diagnostic readings depending on how the dynamic range is mapped—underscoring that the detector records a single quantitative measurement, and the clinical image is a downstream choice.

Conventional CT collapses an energy-dependent attenuation profile into a single number, discarding the spectral signature that distinguishes materials of equal density but different composition. Spectral imaging recovers that information by sampling attenuation at two or more effective energies, exploiting the fact that, in the diagnostic range, $\mu(E)$ for any material is well approximated as a weighted sum of two energy-dependent basis functions—one tracking the photoelectric effect, one tracking Compton scatter: $\mu(E) \approx a_{\mathrm{PE}}\, f_{\mathrm{PE}}(E) + a_{\mathrm{C}}\, f_{\mathrm{C}}(E)$. Because there are effectively two physical processes, two independent energy measurements suffice to solve for two basis coefficients, and from them for the densities of two chosen basis materials (commonly water and iodine, or water and calcium). This is the mathematical engine of material decomposition.

Dual-energy CT (DECT) acquires the two spectra by one of several engineering strategies: rapid kV-switching within a single tube, two tubes at different potentials on the same gantry (dual-source), a layered “sandwich” detector that separates low- and high-energy photons by depth, or split-filtration of one beam. Each trades temporal coregistration, spectral separation, and dose differently, but all feed the same decomposition. From the solved coefficients the scanner synthesizes clinically powerful derived images. Virtual monoenergetic images (VMI) reconstruct the volume as if acquired with a monochromatic beam at a chosen keV; low-keV VMIs (40–55 keV) exploit the steep low-energy photoelectric rise to amplify iodine contrast for subtle hypervascular lesions or poorly opacified vessels, while high-keV VMIs ($\ge 100\,\mathrm{keV}$) suppress beam-hardening and metal streak by moving away from the photoelectric regime. Iodine maps quantify contrast distribution directly, separating true enhancement from intrinsic hyperdensity and enabling perfused-blood-volume imaging in pulmonary embolism. Virtual non-contrast (VNC) images subtract the iodine basis to reconstruct an unenhanced-appearing dataset from a single contrast acquisition, eliminating a separate non-enhanced pass and its dose. Effective-$Z$ maps colorize composition, the classic application being confident discrimination of uric acid from calcium-containing urinary stones—a decision that changes management from lithotripsy to alkalinization and is essentially impossible on single-energy HU alone.

Photon-counting CT (PCCT), the most consequential recent advance, replaces the energy-integrating scintillator with a direct-conversion semiconductor (cadmium telluride or cadmium zinc telluride) in which each absorbed photon generates an electron–hole cloud whose charge pulse is registered individually. Pulse-height analysis sorts every photon into multiple energy bins, so spectral information is captured intrinsically at the detector in a single acquisition rather than reconstructed from dual spectra. The physical advantages are substantial: electronic noise is rejected by setting an energy threshold below which counts are ignored, eliminating the electronic-noise floor that plagues low-dose energy-integrating systems; equal weighting (or optimal energy weighting) of low-energy photons improves iodine contrast-to-noise and dose efficiency; the absence of optical septa and the small native detector pitch permit markedly higher spatial resolution; and multi-energy binning enables richer, multi-material decomposition and even quantitative K-edge imaging of high-$Z$ contrast agents. PCCT generalizes the two-material framework to several materials simultaneously, approaching the long-standing goal of compositional imaging.

The expert reasoning is to choose the spectral product to the question and to remain alert to its limits. Quantitative iodine and effective-$Z$ values depend on calibration, patient size, and decomposition assumptions; VNC slightly mis-estimates true unenhanced HU and can under-call small calcifications; low-keV VMIs amplify noise as they amplify contrast. The cognitive failure mode is to treat a derived map as ground truth rather than as a model-based estimate. Used with that discipline, spectral imaging reframes CT from a density map into a quantitative compositional instrument—reducing equivocal findings, removing entire acquisitions and their dose, and resolving differentials that single-energy attenuation cannot.