COMPUTED TOMOGRAPHY Physical Principles, Clinical Applications, and Quality Control FOURTH EDITION

If you touch a CT scanner—as a student, technologist, or radiographer preparing for boards—this is the physics book that explains rather than intimidates. Seeram connects every knob you turn to a principle: mAs to noise, pitch to coverage, kernel to sharpness. Even curious patients come away understanding what those '0 ± 5 HU' numbers mean. It rewards anyone who wants to defend their imaging decisions with confidence.

1. 1Chapter 1 / Appendix A

   Nine days for one picture: the bet that paid a Nobel Prize

   At Atkinson Morley's Hospital in Wimbledon, London, in 1971, a patient with a suspected brain tumor lay still while Hounsfield's prototype gathered data—a process that had taken about nine days per image in early bench tests, with the computer needing 2.5 hours to reconstruct a single slice. The scan worked: the lesion showed clearly, and the results were presented at the 1972 British Institute of Radiology meeting. The original EMI scanner imaged the head only, coupled through a water bag, on an 80 x 80 matrix, taking about 5 minutes per scan and 20 minutes to reconstruct, each voxel a 3 mm x 3 mm x 13 mm block. Hounsfield and Allan Cormack later shared the 1979 Nobel Prize in Physiology or Medicine for computer-assisted tomography.

   The idea

   The surprise is how slow the first proof of concept was—nine days of acquisition for one image—yet the principle held: projection measurements alone could rebuild a picture of the interior. The first clinical CT scan was performed in 1971 at Atkinson Morley's Hospital in Wimbledon, London, on a patient with a suspected brain tumor. The water bag, the coarse matrix, the 20-minute waits all vanished in later decades, but the core idea—measure attenuation from many angles, then compute—never changed.

   Try this: Next time you review an old head CT, note the matrix size and slice thickness in the header—then compare with today's protocol to see how far voxel sizes have shrunk since 3 x 3 x 13 mm.

2. 2Chapter 1 / Chapter 2

   Water is zero: the scale that names every shadow

   Behind every CT image sits a measurement of the linear attenuation coefficient μ, computed voxel by voxel from transmitted intensities using the Beer-Lambert law, I = I0 * e^(-mu * x). The original EMI scanner's attenuation values were scaled so water equaled 0 and air equaled -1000, later standardized as Hounsfield units. On that scale, dense bone reaches about +1000 or higher, and every tissue finds its place between air and bone.

   The idea

   Anchoring the scale to water made scanner outputs comparable across machines and decades, because water is stable and universally available. The formula HU = 1000 * (mu - mu_water) / mu_water means the number you read is a relative comparison, not an absolute density.

   Try this: Put a region of interest on water (a cyst, the bladder, or your daily phantom) and confirm it reads 0 ± 5 HU before trusting any other measurement.

3. 3Chapter 2 / Chapter 3 / Chapter 4

   Filter first, then smear: beating the star artifact

   Iterative reconstruction (IR) methods, including the algebraic reconstruction technique (ART), were used in early CT but abandoned in favor of filtered backprojection (FBP) due to computational demands. Backprojection alone paints every projection across the image matrix and produces a star-like blurring artifact; the fix is to convolve each projection with a filter kernel—the Ram-Lak filter among them—before backprojecting. The convolution theorem explains why this works: convolution in the spatial domain equals multiplication in the frequency domain, so a sharp or smooth kernel chosen in frequency space shapes the image directly.

   The idea

   The elegance is that filtering happens on the raw data, one dimension at a time, before the computationally heavy backprojection step—cheap math that prevents expensive ugliness. Choosing a kernel is a clinical decision disguised as a physics one: sharp kernels preserve detail but amplify noise, smooth kernels calm the image but soften edges. IR algorithms are now increasingly used to reduce noise and dose.

   Try this: Reconstruct the same dataset with a sharp and a smooth kernel and compare noise in a uniform region—see the trade-off with your own eyes.

4. 4Chapter 2 / Chapter 3

   From translate-rotate to one-second spins: four generations

   First and second generation scanners use rotate-translate geometry, building one slice step by step. Third generation changed the geometry entirely—x-ray tube and detector array locked together in a fan beam, rotating around the patient as one unit. That rotate-rotate design cut a full 360-degree rotation to about a second for a single slice, a dramatic drop from the rotate-translate crawl. Fourth generation uses stationary-detector rotate geometry, with detectors in a stationary ring while only the tube rotates.

   The idea

   Each generation solved the bottleneck of the last—first detector coverage, then cable wind-up, then speed.

   Try this: When you next see a gantry spec sheet, find the rotation time and multiply by how many slices per rotation—feel how generation history compressed into one number.

5. 5Chapter 3 / Chapter 5

   Cutting the cable: slip rings set the data free

   Slip-ring technology enabled continuous (spiral) scanning by removing the need for cables between the rotating and stationary portions of the gantry, using brushes contacting rotating rings to transmit power and data. The gantry no longer has to reverse. It can rotate continuously, and once it does, the table can feed steadily through a beam that never stops spinning. That is spiral, or helical, scanning.

   The idea

   Helical scanning is often credited to detectors or computers, but the enabling hardware was electrical: a mechanical connection problem solved with sliding contacts.

   Try this: Trace a helical dataset's pitch parameter back to the gantry's continuous rotation and explain to a student why the table never stops mid-scan.

6. 6Chapter 3 / Chapter 9

   Pitch: how far the table dares to travel

   In spiral CT, pitch is table feed per rotation divided by slice thickness—or, on multidetector machines, by the total collimation width. A pitch greater than 1 means the table moves farther than the beam width per rotation.

   The idea

   Pitch is defined as table feed per rotation divided by slice thickness (or beam collimation collimation width).

   Try this: Check a helical chest protocol's pitch, compute table feed per rotation from total collimation, and predict whether gaps or overlap exist.

7. 7Chapter 6 / Chapter 9 / Chapter 10

   Many rows, one heartbeat: the MDCT leap

   Multidetector-row CT put arrays of detector rows—4, then 16, then 64, then 320 slices—acquiring several slices per rotation, reducing scan time, improving z-axis resolution, and dramatically increasing speed and volume coverage. But the wider cone of x-rays brought a new problem: rays no longer confined to a perfect plane, causing artifacts at the edges of the detector rows. Specialized cone-beam reconstruction algorithms, such as the Feldkamp-Davis-Kress (FDK) method—which approximates the cone beam as a series of fan beams—were needed to correct them.

   The idea

   The more rows you add, the more the cone-beam geometry departs from planar assumptions, and FDK's approximation—treating a cone as stacked fans—addresses this.

   Try this: On a wide-cone scanner, image a uniform phantom and look for shading near the z-axis edges—evidence of cone-beam effects the algorithm only partly tames.

8. 8Chapter 6

   Slicing the body sideways: MPR and 3D

   MPR is a post-processing technique that uses axial slice data to generate images in planes other than the axial plane, such as coronal, sagittal, or oblique images. Three-dimensional surface rendering is also introduced as a post-processing technique.

   The idea

   MPR and 3D surface rendering are post-processing techniques that work from existing axial slice data.

   Try this: Take a routine abdomen dataset and build a coronal reformation through the kidneys—notice how much easier anatomy is to explain to a patient that way.

9. 9Chapter 7

   Line pairs and low contrast: what 'good image' means

   Spatial resolution is the ability to distinguish small, high-contrast objects placed close together, measured with line-pair phantoms or through point-spread and modulation-transfer functions. Contrast resolution is a different skill: distinguishing attenuation differences between tissues of similar densities, improved by increasing slice thickness, increasing mAs, and using larger pixels—and superior in CT compared to conventional radiography. Noise, defined as the random variation (standard deviation) of CT numbers across a uniform region of interest, is inversely proportional to the square root of the number of photons detected; raising mAs reduces noise but raises patient dose.

   The idea

   Spatial resolution, contrast resolution, and noise are distinct image-quality measures, each with its own measurement method and its own trade-offs.

   Try this: Measure the standard deviation in a uniform region on your next QC scan, then halve the mAs on a phantom and watch the noise rise by roughly the square root of two.

10. 10Chapter 7 / Chapter 8

    When the image lies: naming the artifacts

    An artifact is any discrepancy between the CT numbers in the reconstructed image and the true attenuation coefficients of the object, and the chapter categorizes artifacts by cause: physics-based ones like beam hardening, partial volume, and quantum noise; patient-based ones like motion and metal; scanner-based ones like ring artifacts from detector miscalibration.

    The idea

    Classifying artifacts by cause—physics, patient, or scanner—points to the appropriate fix.

    Try this: Pull up your last image with a ring artifact, log it, and check the detector calibration history—connect the visual pattern to its scanner-based cause.

11. 11Chapter 5 / Chapter 8 / Chapter 14

    Counting the dose: CTDI, DLP, and the 100 mm pencil

    Radiation dose in CT is tracked through descriptors: the CT dose index (CTDI), the dose-length product (DLP), and the multiple scan average dose (MSAD). CTDI_100 is measured with a 100 mm long pencil ionization chamber in standard phantoms—a 16-cm acrylic phantom for head scans, a 32-cm phantom for body scans—and the weighted CTDI combines center and periphery as (1/3) * CTDI100(center) + (2/3) * CTDI100(periphery). DLP then multiplies CTDI_vol by scan length, in units of mGy·cm, reflecting the total radiation energy imparted for a scan series.

    The idea

    The weighted formula's 1/3 and 2/3 split reflects that peripheral measurements count more heavily than central ones. CTDI describes a single slice's intensity; DLP adds the length dimension.

    Try this: Find the DLP printed on your next study's dose report and divide by scan length to recover CTDI_vol—verify the machine's own arithmetic.

12. 12Chapter 11 / Chapter 14

    Children first: ALARA as a promise, not a slogan

    Children's greater radiation sensitivity and longer life expectancy for risk manifestation make pediatric CT dose optimization especially important. The book centers dose optimization on the ALARA principle—As Low As Reasonably Achievable—meaning every parameter is chosen to answer the clinical question with the least radiation that still produces a diagnostic image.

    The idea

    ALARA reframes dose reduction as a principle: the goal is the sufficient image, not the prettiest one.

    Try this: Review one pediatric protocol this week and check whether mAs, kVp, and scan length are actually scaled to patient size rather than copied from adult defaults.

13. 13Chapter 13 / Chapter 14

    Timing the bolus: CTA and the needle's eye view

    CTA requires precise timing of contrast injection, often using a bolus-tracking technique with a region of interest in the ascending aorta to time the contrast injection for optimal arterial opacification. Multiphase liver CT protocols include non-contrast, arterial phase (~25-30 sec delay), and portal venous phase (~60-70 sec delay) for characterizing lesions like hemangiomas or hepatocellular carcinoma.

    The idea

    Contrast-enhanced CT depends on timing: tissues enhance at different rates, and the phase captured determines what the lesion reveals.

    Try this: Watch a bolus-tracking scan run and note the ROI's position in the ascending aorta—then check how the trigger threshold shaped the arterial phase image.

14. 14Chapter 8 / Chapter 15 / Chapter 17

    The phantom that keeps you honest: QC and the ACR test

    A uniform water phantom is scanned and the mean CT number should be 0 ± 5 HU, with the standard deviation (noise) within acceptable limits, and deviations indicate calibration problems. Checks of alignment, slice thickness, and table increment accuracy are also routine. For comprehensive testing, the American College of Radiology (ACR) CT accreditation phantom carries modules for CT number accuracy, low-contrast resolution, high-contrast resolution, uniformity, slice thickness, and radiation dose.

    The idea

    Quality control verifies that the scanner's numbers remain trustworthy: the water phantom anchors the HU scale, and the ACR phantom bundles image-quality tests into one object.

    Try this: Run or review today's daily water phantom check and confirm the mean sits within 0 ± 5 HU before the first patient of the day.

15. 15Chapter 11 / Chapter 12 / Chapter 13

    Where CT earns its keep: head, chest, bone

    Non-contrast head CT is used for acute trauma, stroke (hemorrhage), and headache, visualizing fractures, hemorrhage, and midline shift. HRCT of the chest is used for interstitial lung disease, using thin slices (1-2 mm) with a high-spatial-frequency reconstruction algorithm to visualize fine lung detail. CT is superior to radiography for evaluating complex fractures, especially in areas like the pelvis, acetabulum, spine, and wrist (e.g., scaphoid fractures); it is the preferred modality for evaluating complex fractures, such as tibial plateau fractures, because it provides detailed 3D information, with the Schatzker classification system used to categorize tibial plateau fractures into six types.

    The idea

    Each application bends the technology to a different virtue: head CT prizes speed and hemorrhage sensitivity in emergencies; HRCT sacrifices coverage for fine lung texture; musculoskeletal CT supplies the 3D geometry needed for complex fractures.

    Try this: Pull one HRCT chest case and confirm the slice thickness is 1-2 mm with a sharp algorithm—then view the same data smoothed to see what fine detail disappears.

Radiography students, CT technologists preparing for certification, and radiology residents who need physics they can actually apply at the console. Also ideal for educators seeking a clear, complete CT curriculum in one volume.

FAQ