# CT Physics: Acquisition, Reconstruction, and Dose

## CT Acquisition Geometry

### Basic Principles

Computed tomography produces cross-sectional images by measuring x-ray attenuation from multiple angles around the patient. An x-ray tube and detector array rotate within a circular gantry while the patient lies on a table at the center. At each angular position, the system collects attenuation data as projections (ray sums). Mathematical reconstruction algorithms then convert this projection data into a two-dimensional image matrix, where each pixel is assigned a CT number expressed in Hounsfield units.

### Hounsfield Units (HU)

The Hounsfield unit scale provides a standardized, quantitative measure of tissue attenuation. It is defined as CT number = 1000 x (mu_tissue - mu_water) / mu_water, where mu represents the linear attenuation coefficient. Water is defined as 0 HU and air as -1000 HU, with dense bone at approximately +1000 HU. Fat typically falls between -50 and -100 HU, and most soft tissues range from +20 to +80 HU. Because CT numbers are linear and quantitative, they enable tissue characterization based on attenuation values, a capability that plain radiography cannot match.

| Tissue | Typical HU Range |
|--------|-----------------|
| Air | -1000 |
| Lung | -500 to -900 |
| Fat | -50 to -100 |
| Water | 0 |
| Soft tissue (muscle, organs) | +20 to +80 |
| Acute blood | +50 to +70 |
| Calcification | +100 to +300 |
| Dense bone | +1000 |

### Helical (Spiral) CT

Modern CT scanners acquire data in helical (spiral) mode, with the x-ray tube rotating continuously while the patient table translates through the gantry simultaneously. The key parameter describing this geometry is pitch, defined as the table travel per rotation divided by the total collimation width. A pitch greater than 1 means there are gaps in the data (faster scan, lower dose, but potentially reduced z-axis resolution), while a pitch less than 1 means the helices overlap (slower scan, higher dose, better z-axis sampling). Interpolation algorithms reconstruct axial images from the helical dataset.

### Multi-Detector CT (MDCT)

Current clinical scanners use multiple rows of detectors (64 to 320 rows), which acquire multiple slices simultaneously during each gantry rotation. This wider z-axis coverage per rotation enables faster scanning and improved temporal resolution, which is especially important for cardiac and trauma imaging. Sub-millimeter detector elements (0.5 to 0.625 mm) allow isotropic voxel acquisition, meaning the voxel dimensions are equal in all three directions. Isotropic voxels are what make high-quality multiplanar reformats and 3D reconstructions possible in any arbitrary plane without loss of resolution.

### Dual-Energy CT (DECT)

Dual-energy CT acquires data at two different x-ray energy spectra, typically 80 and 140 kVp. This exploits the fact that different materials attenuate x-rays differently depending on the photon energy. Several hardware implementations exist, including dual-source systems, rapid kVp switching, dual-layer (sandwich) detectors, and twin-beam filtration. Clinically, DECT enables applications such as differentiating uric acid from calcium stones, generating virtual non-contrast images from contrast-enhanced datasets (potentially eliminating a scan phase), quantifying iodine in tissues, reducing metal artifacts, and detecting monosodium urate crystals in gout.

## Image Reconstruction

### Filtered Back Projection (FBP)

Filtered back projection has been the standard reconstruction algorithm since the earliest days of CT. Each projection is filtered through convolution with a mathematical kernel to remove blurring, then back-projected across the image matrix. FBP is computationally fast but produces more noise at low radiation doses. The choice of reconstruction kernel (also called a filter) determines the tradeoff between noise and spatial resolution: smooth (soft tissue) kernels produce lower noise but lower spatial resolution, while sharp (bone or lung) kernels yield higher spatial resolution at the cost of more noise.

### Iterative Reconstruction (IR)

Iterative reconstruction algorithms represent a significant advance over FBP. They use statistical modeling and iterative correction cycles to reduce image noise, enabling equivalent image quality at substantially lower radiation doses, with potential dose reductions of 20 to 50%. Several types exist, ranging from statistical approaches (such as ASIR and SAFIRE) to fully model-based methods (such as MBIR/Veo, ADMIRE, and IMR). Model-based IR produces the lowest noise but can alter image texture, giving images a "waxy" or "plastic" appearance at aggressive settings that some radiologists find unfamiliar. Hybrid approaches that blend IR with FBP help maintain the familiar image texture that radiologists are accustomed to reading.

| Reconstruction Method | Mechanism | Dose Reduction | Image Texture | Examples |
|----------------------|-----------|---------------|---------------|----------|
| Filtered Back Projection (FBP) | Convolution + back projection | Baseline | Familiar, granular | Standard (all vendors) |
| Hybrid Iterative (IR) | Statistical modeling + FBP blend | 20-40% | Mostly preserved | ASIR (GE), SAFIRE (Siemens) |
| Model-Based IR | Full physics modeling, iterative cycles | 40-50% | May appear "waxy" | MBIR/Veo (GE), ADMIRE (Siemens), IMR (Philips) |
| AI / Deep Learning | Trained neural networks | ≥50% potential | Well-preserved | TrueFidelity (GE), AiCE (Canon), Precise Image (Philips) |

### Artificial Intelligence (AI) / Deep Learning Reconstruction

The newest generation of reconstruction methods uses trained neural networks to reduce noise. These AI-based approaches can provide noise reduction superior to iterative reconstruction without altering image texture in the same way, and they hold promise for further dose reduction beyond what IR alone achieves. Examples include TrueFidelity (GE), AiCE (Canon), and Precise Image (Philips).

## CT Image Quality

### Spatial Resolution

In-plane spatial resolution in CT ranges from 0.5 to 1.0 mm, depending on the reconstruction kernel and field of view. Z-axis resolution is determined by detector element width and reconstructed slice thickness. High-resolution kernels designed for bone and lung imaging improve spatial resolution but increase noise.

### Contrast Resolution

Contrast resolution, the ability to distinguish structures with small differences in attenuation, is where CT truly excels compared to plain radiography. However, it is limited by noise: improving contrast resolution requires increasing the radiation dose. Low-contrast detectability is the primary reason CT is so much more diagnostically powerful than radiography for soft tissue evaluation.

### Noise

Quantum noise in CT is proportional to 1/sqrt(mAs) and inversely related to slice thickness. Doubling the mAs reduces noise by approximately 30% (a factor of 1/sqrt(2)). Thinner slices produce noisier images than thicker slices at the same mAs because fewer photons contribute to each slice. Patient size is also a major factor: larger patients attenuate more photons before they reach the detector, which increases noise in the resulting image.

### Artifacts

CT images are susceptible to several types of artifacts. Beam hardening occurs because the polychromatic x-ray beam becomes progressively harder (higher mean energy) as it passes through tissue, producing cupping artifacts in uniform objects and streaks between dense structures like the petrous bones or dental hardware. Photon starvation causes increased noise and streaking through highly attenuating regions like the shoulders and pelvis, where few photons make it through to the detectors. Motion artifacts appear as blurring and streaking from patient or organ movement. Ring artifacts result from miscalibrated individual detector elements and appear as concentric rings in the image. Metal artifacts produce severe streaking adjacent to metallic implants and can be managed with metal artifact reduction (MAR) algorithms and dual-energy CT techniques. Partial volume averaging occurs when different tissues are mixed within a single voxel, leading to inaccurate CT numbers that represent an average rather than any single tissue. Cone-beam artifacts cause geometric distortion at the periphery of wide-area detectors.

## CT Dose Metrics

### CTDIvol (Volume CT Dose Index)

CTDIvol represents the average radiation dose within a standardized cylindrical phantom for a single tube rotation. It is measured in milligray (mGy) and is standardized to either a 16 cm phantom (for head protocols) or a 32 cm phantom (for body protocols). It is important to understand that CTDIvol is a scanner output metric, not a measure of actual patient dose. Its primary purpose is to allow standardized comparison between scanners and protocols, and it is displayed on the scanner console before each scan.

### DLP (Dose-Length Product)

The dose-length product, calculated as CTDIvol multiplied by the scan length in centimeters, is measured in mGy-cm. Because it accounts for the total extent of tissue irradiated, DLP is a better indicator of total patient radiation exposure than CTDIvol alone.

### Effective Dose

Effective dose is an estimated whole-body dose that reflects the weighted sum of doses to individual organs, accounting for the varying radiation sensitivity of different tissues. It is calculated as effective dose (mSv) = DLP x k, where k is a conversion factor that depends on the body region scanned. Typical k-factors are approximately 0.0021 for the head, 0.014 for the chest, and 0.015 for the abdomen and pelvis. This yields typical effective doses of about 2 mSv for a head CT, 5 to 7 mSv for a chest CT, and 8 to 10 mSv for an abdomen/pelvis CT. Effective dose is useful for comparing radiation risk across different imaging modalities.

| CT Examination | k-Factor (mSv/mGy·cm) | Typical Effective Dose (mSv) |
|---------------|------------------------|------------------------------|
| Head CT | 0.0021 | 1-2 |
| Chest CT | 0.014 | 5-7 |
| Abdomen/Pelvis CT | 0.015 | 8-10 |
| Coronary CTA | 0.014 | 3-5 |
| Annual background radiation | — | ~3 |

### Size-Specific Dose Estimate (SSDE)

SSDE adjusts the CTDIvol based on the actual size of the patient, using either the effective diameter or water-equivalent diameter. This makes it a better approximation of the dose actually received by an individual patient, which is especially important in pediatric and small-adult imaging where the standard 32 cm phantom significantly overestimates dose. SSDE is calculated as CTDIvol multiplied by a size-specific conversion factor derived from AAPM Reports 204 and 220.

## Dose Optimization Strategies

### Technique Optimization

Several technical strategies can reduce radiation dose. Automatic tube current modulation (ATCM) adjusts the mA in real time based on patient attenuation, both angularly (adjusting for the difference between the AP and lateral dimensions of the body) and along the z-axis. Automatic kVp selection lowers the tube voltage for smaller patients and for contrast-enhanced studies, which improves iodine conspicuity while reducing dose. Limiting the scan range to match the clinical question and selecting only the necessary contrast phases (avoiding unnecessary multiphase acquisitions) are simple but effective dose reduction measures.

### Reconstruction-Based Dose Reduction

Iterative reconstruction enables 20 to 50% dose reduction while maintaining diagnostic image quality. AI-based reconstruction may permit further reductions. When thin-slice detail is not needed for the clinical question, reconstructing thicker slices for interpretation reduces noise and can allow the use of lower-dose acquisition protocols.

### Protocol Design

Standardized, indication-based protocols aligned with the ACR Appropriateness Criteria ensure that each scan is tailored to the clinical question. Size-based pediatric protocols, championed by the Image Gently campaign, are essential for the youngest patients. Diagnostic reference levels (DRLs) allow institutions to benchmark their dose performance against national standards, identifying opportunities for optimization.

<image>A diagram showing helical CT acquisition geometry. The x-ray tube traces a helical path around the patient as the table moves continuously through the gantry. The diagram shows the relationship between table feed per rotation and detector collimation width, defining pitch. Three examples are illustrated: pitch less than 1 (overlapping helices), pitch equal to 1 (contiguous), and pitch greater than 1 (gaps requiring interpolation). The multi-detector array is shown with multiple rows of detector elements capturing multiple slices per rotation.</image>

<image>A comparison of CT image quality using three reconstruction methods on the same low-dose abdominal CT dataset. Left panel: filtered back projection (FBP) showing significant quantum noise and grainy texture. Middle panel: hybrid iterative reconstruction showing moderate noise reduction with preserved anatomic detail. Right panel: model-based iterative reconstruction showing the lowest noise but slightly altered image texture with a smoother, less granular appearance. All three images show the same axial slice at the level of the liver. Dose and noise measurements (CTDIvol and standard deviation of HU in a homogeneous region) are annotated on each image.</image>

<image>An infographic illustrating CT dose metrics. A cylindrical PMMA phantom (16 cm head, 32 cm body) is shown with a pencil ionization chamber inserted at center and periphery positions for CTDI measurement. Arrows connect CTDIvol to DLP (CTDIvol multiplied by scan length) and then to effective dose (DLP multiplied by a region-specific conversion factor k). A sidebar table shows typical effective doses for common CT examinations: head CT 1-2 mSv, chest CT 5-7 mSv, abdomen/pelvis CT 8-10 mSv, coronary CTA 3-5 mSv, compared to annual background radiation of approximately 3 mSv.</image>

## Clinical Pearls

CTDIvol is a scanner output metric, not a patient dose; SSDE adjusts for patient size and provides a better estimate of individual dose, which is especially important in children. Every CT dose report (CTDIvol and DLP) should be reviewed to catch protocol errors or unusually high doses. Iterative reconstruction can reduce dose by 20 to 50%, but aggressive settings may alter image texture and reduce diagnostic confidence, so each institution must find its own balance. The linear no-threshold (LNT) model remains controversial; current evidence does not conclusively demonstrate cancer risk at diagnostic CT dose levels, but the ALARA principle remains prudent. Dual-energy CT has moved beyond being a physics curiosity and now has practical clinical applications, including virtual non-contrast images, gout detection, and improved characterization of renal stones and adrenal lesions. Automatic tube current modulation significantly reduces dose in asymmetric body regions such as the shoulders and pelvis, but it depends on correct patient centering in the gantry to function properly.

## References

- Bushberg JT, et al. *The Essential Physics of Medical Imaging*, 4th edition
- AAPM Report No. 204: "Size-Specific Dose Estimates (SSDE) in Pediatric and Adult Body CT Examinations"
- AAPM Report No. 220: "Use of Water Equivalent Diameter for Calculating Patient Size and Size-Specific Dose Estimates (SSDE) in CT"
- McCollough CH, et al. "CT dose: how to measure, how to reduce." *RadioGraphics*, 2011
- Geyer LL, et al. "State of the art: iterative CT reconstruction techniques." *Radiology*, 2015
- ACR-AAPM Practice Parameter for Diagnostic Reference Levels and Achievable Doses in Medical X-Ray Imaging