3. Image Acquisition

The geometry of data acquisition determines what a CT scanner can and cannot resolve, and every modern protocol is an exercise in exploiting—or working around—the constraints of that geometry. In the third-generation rotate–rotate configuration that dominates clinical practice, a fan- or cone-shaped polychromatic x-ray beam is emitted from a focal spot on the anode and intercepted by an arc of detector elements rigidly coupled to the tube, the entire assembly rotating about the patient on a slip-ring gantry. Each projection samples the line integral of the linear attenuation coefficient along ray paths, $p(\theta, s) = \int_{L(\theta,s)} \mu(x,y)\,dl$, and the reconstruction problem is the inversion of this Radon transform. The fidelity of that inversion is gated entirely by how completely and how finely the $(\theta, s)$ space is sampled, which is precisely what the acquisition mode controls.

In axial (step-and-shoot) acquisition, the table is stationary while the gantry completes one or more rotations, a contiguous set of slices is read out, the table indexes by a fixed increment, and the cycle repeats. Because no $z$-axis motion contaminates a given rotation, axial mode yields the cleanest in-plane data and the most predictable section sensitivity profile; it remains the reference standard for noncontrast head CT, where the posterior fossa is exquisitely vulnerable to the interpolation streaks of helical scanning, and for high-resolution temporal-bone and quantitative tasks (coronary calcium scoring, perfusion at a fixed slab) where temporal registration and an undistorted slice profile matter more than speed. Its liabilities are inter-scan delay, susceptibility to misregistration from patient motion between steps, and an inability to freeze a contrast bolus across a long $z$-extent.

Helical (spiral) acquisition removes the inter-scan dead time by translating the table continuously during gantry rotation, so the focal spot traces a helix relative to the patient. No single rotation now lies in a plane, so the raw data must be longitudinally interpolated to synthesize the planar projection set required at each reconstruction position $z_0$; classically this is 180°LI or 360°LI interpolation, in which conjugate rays half a rotation apart are weighted by their $z$-distance from $z_0$. This interpolation is the source of helical-specific artifacts—the pinwheel/windmill pattern at high-contrast edges and stair-step distortion of obliquely oriented structures—but it is also what enables volumetric coverage of an entire organ within a single breath-hold, the prerequisite for CT angiography and multiphase imaging.

Multidetector (MDCT) acquisition generalizes the helix into the third dimension by replacing the single detector row with an array of $N$ rows, so that one rotation simultaneously samples a slab of thickness $N \cdot T$ (detector rows of width $T$). Coverage per rotation scales with $N$, collapsing breath-holds and enabling isotropic voxels reconstructable in any plane. The cost is that the beam diverges as a cone rather than a fan; rays no longer lie in transaxial planes, and exact 2D filtered back projection is invalid. Cone-beam reconstruction (approximate Feldkamp-class or advanced exact/iterative algorithms) becomes mandatory, and the residual cone-angle error grows toward the periphery of wide ($\geq$160-mm) detectors, manifesting as cone-beam shading and intensity drop-off. Wide-array "volume" scanners can image an entire organ in a single non-helical rotation, eliminating $z$-interpolation artifacts and freezing the bolus for true 4D perfusion—at the price of the steepest cone-angle penalty in CT.

A handful of operator-set parameters fully specify a helical acquisition, and an expert reads them the way a spectroscopist reads instrument settings: each one trades resolution, noise, dose, and time against the others through relationships that are quantitative, not heuristic.

Pitch is the dimensionless ratio of table travel per gantry rotation to total nominal beam width, $p = \dfrac{d_{table}}{N \cdot T}$. At $p=1$ the helix is contiguous; at $p>1$ the helix is stretched and a given voxel is irradiated by a shorter arc, so for fixed tube current the dose falls in proportion, $\text{CTDI}_{vol} = \dfrac{\text{CTDI}_{w}}{p}$, while image noise rises as fewer photons contribute per reconstructed plane. This is why most vendors expose an effective mAs, $\mathrm{mAs}_{eff} = \dfrac{\mathrm{mAs}}{p}$, and hold it constant so that increasing pitch shortens scan time without silently degrading noise. High pitch ($p \approx 1.2\text{–}1.5$, or $>3$ on dual-source systems) freezes motion and bolus but widens the effective section and can amplify windmill artifact; low pitch ($p<1$) oversamples for thin-section and cardiac work at a dose premium. The temporal-resolution and z-coverage consequences are inseparable from pitch, which is why it is the single most clinically loaded number in the protocol.

Collimation must be parsed at three levels that novices conflate. The detector configuration ($N \times T$, e.g. $128 \times 0.6$ mm) sets the thinnest achievable section and the data channels. The nominal beam collimation ($N \cdot T$) sets z-coverage per rotation. Neither equals the reconstructed slice thickness, which can be made thicker than $T$ but never reliably thinner. Two dose penalties live here. Overbeaming: the focal spot is finite, so the umbral plateau of the dose profile must exceed the active detector to keep edge rows in full flux; the wasted penumbra is a fixed z-width, so its fractional dose cost is large for narrow arrays and negligible for wide ones. Overranging (z-overscanning): helical interpolation needs data half a rotation beyond each end of the planned range, adding $\approx p \cdot N \cdot T$ of extra-irradiated length—again punitive for wide collimation and high pitch, and the rationale for adaptive z-collimators that asymmetrically shutter the leading/trailing edges.

Rotation time $t_{rot}$ governs temporal resolution. A standard half-scan (180° + fan angle) reconstruction achieves an effective temporal window of roughly $t_{rot}/2$; at $t_{rot}=0.25$ s this is $\approx125$ ms per source, halved again to $\approx65\text{–}75$ ms by dual-source geometry, the threshold for reliable coronary imaging without strict rate control. Multisector (segmented) reconstruction stitches projections from several heartbeats to push the window lower, trading temporal resolution for sensitivity to beat-to-beat variability—a classic failure mode producing banding artifact at sector boundaries. Faster rotation also reduces tube-loading dwell, capping achievable mAs.

Slice thickness is set by the z-axis point-spread function: thinner sections sharpen the slice-sensitivity profile and minimize partial-volume averaging, in which a voxel spanning two tissues reports the volume-weighted mean attenuation, $\mu_{voxel}=\sum_i f_i \mu_i$, blurring small or obliquely oriented structures and faking lesions at interfaces. The penalty is noise: halving thickness halves the photons per voxel, so $\sigma \propto 1/\sqrt{\text{thickness}}$. The expert resolution is to acquire thin (sub-millimetre, isotropic) and reconstruct task-matched thick sections for review while retaining thins for MPR and problem-solving.

Protocol design is applied physiology: the acquisition is timed and weighted so that the tissue contrast generated by a disease mechanism falls within the acquired window. Each body region imposes its own constraints.

Neuro. Noncontrast head CT prioritizes low-contrast detectability of grey–white differentiation (a $\sim$8–10 HU difference) over spatial resolution, so it uses axial mode or low-pitch helical, thick (4–5 mm) sections to suppress noise, a smooth kernel, and narrow brain windowing ($WW\approx80,\,WL\approx40$); a complementary stroke window ($WW\approx30\text{–}40$) accentuates early ischaemic hypoattenuation. Posterior-fossa beam-hardening (Hounsfield dark-band artifact) is the dominant failure mode and a reason to retain axial acquisition there.

Chest. High-resolution lung assessment exploits the enormous intrinsic air–tissue contrast, permitting thin ($\leq1$ mm) isotropic sections, a sharp/edge-enhancing kernel, and lung windowing ($WW\approx1500,\,WL\approx-600$) at low dose; a soft kernel and mediastinal window ($WW\approx350,\,WL\approx50$) are reconstructed from the same data for nodes and vessels. Inspiratory breath-hold is mandatory to avoid dependent atelectasis masquerading as ground-glass; expiratory acquisition is added deliberately to unmask air-trapping.

Abdomen. Soft-tissue lesion conspicuity depends on enhancement kinetics, so abdominal CT is multiphasic. The relevant table:

Phase	Delay after injection	Primary target
Late arterial	$\approx 35\text{–}40$ s	Hypervascular HCC/NET, arterial map
Portal venous	$\approx 65\text{–}80$ s	Hepatic parenchyma, bowel, most metastases
Nephrographic	$\approx 100$ s	Renal masses
Delayed/excretory	$3\text{–}10$ min	Cholangiocarcinoma washout, urothelium

A hypervascular lesion seen only in the late-arterial window is invisible portal-venously—a mistiming miss, not an absent lesion.

Trauma. Whole-body (pan-scan) CT favours speed and completeness: high-pitch helical from vertex to pelvis, split-bolus or arterial-plus-portal venous timing to capture active extravasation (a focal hyperdense blush isodense to aorta) and solid-organ laceration in one pass, with thin isotropic data for multiplanar bone and vascular review. The failure mode is satisfaction of search—fixating on an obvious injury and missing a second.

CTA. Arterial-phase opacification requires synchronizing the scan to peak intraluminal iodine. Bolus tracking places a region of interest in the target artery and triggers when attenuation crosses a threshold (commonly $\sim$100–150 HU); a test bolus instead measures time-to-peak directly. The governing quantity is iodine delivery rate (injection rate $\times$ concentration); scan direction should follow bolus transit, and low kV (70–90 kVp) is exploited because the photoelectric attenuation of iodine rises steeply near its k-edge (33.2 keV), boosting vascular contrast per unit dose. Mistiming yields a falsely "occluded" or poorly opacified segment.

CT perfusion. Perfusion is a dynamic acquisition: the same slab is scanned repeatedly (every 1–3 s for $\sim$60 s) to sample the time–attenuation curve of contrast through tissue. Deconvolution of tissue against an arterial input function yields CBF, CBV, MTT and $T_{max}$; in stroke, an ischaemic core ($\text{CBF}<30\%$ of contralateral) is distinguished from salvageable penumbra ($T_{max}>6$ s mismatch), the physiological basis of late-window thrombectomy selection. The dominant artifacts are patient motion across the time series and a poorly chosen or partial-volumed arterial input, both of which corrupt the deconvolution.