Siemens CT Reconstruction Kernels Decoded

A CT reconstruction kernel is the spatial-frequency filter that turns raw projection data into an image, and on Siemens SOMATOM scanners the kernel you select sets the balance between sharpness, noise, and quantitative accuracy. A smooth kernel suppresses high spatial frequencies to lower noise; a sharp kernel boosts those frequencies to resolve fine detail at the cost of more noise. Pick the wrong one and a clean study becomes unreadable or a calcium score drifts out of tolerance; pick the right one and you get exactly the detail the clinical task needs at the lowest practical dose.¹⁶⁸

This guide decodes the Siemens kernel naming convention, walks through the major kernel families, explains how the underlying reconstruction algorithm — filtered back projection (FBP) versus iterative reconstruction — interacts with kernel choice, and shows how physicists quantify the resolution-versus-noise tradeoff using the modulation transfer function (MTF) and noise power spectrum (NPS). Diagnostic Radiation Physics Services (DRPS) supports CT programs across Florida, Maryland, Virginia, Washington, DC, California, and Nevada, where consistent kernel selection keeps image quality and quantitative measurements audit-ready.

Introduction

The reconstruction kernel is the single protocol parameter that most directly trades spatial resolution against image noise, and on Siemens systems it is encoded in a compact, decodable name. This Physics Pulse guide decodes the Siemens kernel naming convention, walks through the major kernel families, and explains how iterative reconstruction and beam hardening correction interact with kernel choice. It also adds the image-science layer — what "sharper" and "noisier" mean quantitatively in terms of the MTF and NPS — so protocol decisions rest on measurable physics rather than appearance alone.⁶⁸ It is written for the technologists, physicists, and imaging administrators who set and defend protocols during accreditation, and who should always consider kernel, dose, and reconstruction method together, a theme we explore further in CT Protocol Optimization: balancing dose, image quality, and compliance.

Topic Explanation: What a Reconstruction Kernel Does

A reconstruction kernel is a frequency-domain filter applied during image reconstruction that determines how spatial frequencies in the projection data are weighted before the image is formed. In plain terms, the kernel decides how much the scanner emphasizes fine edges versus how much it smooths the image.

• Smooth (soft) kernels suppress high spatial frequencies. They lower image noise and improve low-contrast detectability, at the cost of edge sharpness. They suit soft-tissue tasks such as liver, brain, and routine abdominal imaging.
• Sharp kernels boost high spatial frequencies. They sharpen edges and improve visualization of fine, high-contrast structures such as bone trabeculae and lung interstitium, but they amplify noise.

Because the kernel reshapes the spatial-frequency content of the image, it does not only change "how the image looks." It also changes measured CT numbers in small or high-contrast structures, which is why kernel choice is a quantitative decision, not just an aesthetic one. Kernel, dose, and reconstruction method should always be considered together, a theme we explore further in CT Protocol Optimization: balancing dose, image quality, and compliance.

Filtered back projection versus iterative reconstruction

The kernel never acts alone — it operates inside a reconstruction algorithm, and the two dominant algorithm classes treat the resolution-noise tradeoff very differently.

• Filtered back projection (FBP) is the classic analytic method. The projection data are convolved with the reconstruction kernel (a "filter") and then back-projected across the image matrix in a single pass. FBP is fast and mathematically transparent, but because the kernel is applied directly to the raw data, spatial resolution and noise are rigidly coupled: any kernel that amplifies high spatial frequencies to sharpen edges amplifies high-frequency noise by the same factor.⁶⁸
• Iterative reconstruction (IR) starts from an estimate of the image, forward-projects it to predict what the detector should have measured, compares that prediction to the actual measured projections, and corrects the estimate — repeating the loop and applying statistical and (in model-based variants) physical models of the system. This decouples noise from resolution: IR can suppress noise while largely preserving the high-frequency response the kernel provides.⁶⁷⁸

The practical consequence is central to modern protocol design: with FBP, a sharper kernel always costs noise; with IR, you can often run a sharper kernel at the same dose, or hold the kernel fixed and lower the dose. Notably, IR also changes the texture of the noise — the NPS shifts toward lower spatial frequencies, giving images a characteristic smoother, sometimes "blotchy," appearance at high iterative strengths even when the noise magnitude is reduced.⁶⁸

Understanding the Siemens Kernel Naming Convention

Siemens CT reconstruction kernels follow a standardized naming format that encodes the clinical application and the resolution level. A typical kernel name has two parts: a two-letter marker and a two-digit resolution index, for example Br40 or Qr59.

Kernel Marker (Two Letters)

The two letters identify the kernel family and subgroup.

• First letter — kernel family:
  • B = Body
  • H = Head
  • Q = Quantitative
  • U = Ultra High Resolution
• Second letter — kernel subgroup:
  • r = Regular
  • f = Fine noise optimization
  • l = Lung optimized
  • v = Vascular optimized
  • c = Crisp edge enhancement
  • p = Pediatric optimization

Examples of valid markers include Br, Hr, Qr, and Ur.

Resolution Index (Two-Digit Number)

The two-digit number sets image sharpness and noise level. Higher numbers produce sharper images with greater spatial resolution but increased image noise. Each increment represents approximately a 4% increase in spatial resolution. Examples include Br40, Hr32, Qr59, and Ur77.

A worked example ties it together:

• Br40 — a standard body kernel with medium resolution, commonly used for routine abdominal and chest imaging.

Reading a kernel name therefore tells you, at a glance, both what the kernel is for (family and subgroup) and how aggressively it sharpens (resolution index). The naming convention is essentially a human-readable shorthand for a kernel's frequency response: the marker tells you which task the kernel was tuned for, and the index tells you how far up the spatial-frequency axis its passband extends.

Key Technical Principles: The Resolution–Noise Tradeoff

Quantifying resolution: the modulation transfer function

Spatial resolution is described quantitatively by the modulation transfer function (MTF), which states how faithfully the system reproduces image contrast as a function of spatial frequency $f$ (in line pairs per centimeter or cycles/mm). At low spatial frequencies (large, coarse structures) the MTF is near 1.0 — contrast passes through almost unchanged. As $f$ increases toward fine detail, the MTF falls toward zero, and the kernel sets how quickly that falloff happens.⁶⁸

A common scalar summary is the limiting resolution, often quoted at the spatial frequency where the MTF drops to 10% (sometimes called $f_{10}$):

$$\text{MTF}(f_{10})=0.10$$

A sharper kernel pushes the entire MTF curve to the right, raising the response at high $f$ and increasing $f_{10}$; a smoother kernel rolls the MTF off earlier, lowering $f_{10}$. For modern non-linear reconstructions whose resolution depends on object contrast and dose, AAPM Task Group 233 recommends measuring a contrast- and dose-specific task transfer function (TTF) rather than assuming a single fixed MTF.⁸

Quantifying noise: standard deviation and the noise power spectrum

Noise has two attributes that matter clinically: its magnitude and its texture. Magnitude is the pixel standard deviation $\sigma$ measured in a uniform region (in Hounsfield units), and it is what drives low-contrast detectability. Texture — how the noise is distributed across spatial frequencies — is captured by the noise power spectrum (NPS), the variance of the noise as a function of $f$. The total noise variance is the integral of the NPS over all frequencies:

$${\sigma }^2=\iint \text{NPS}(f_x,f_y)d{f}_{x}d{f}_y$$

The kernel shapes the NPS just as it shapes the MTF. A sharp kernel that boosts high-frequency signal also boosts high-frequency noise, shifting NPS power toward higher $f$ (a "grainier" texture) and raising $\sigma$. A smooth kernel attenuates high-frequency content, lowering $\sigma$ and shifting the NPS peak toward lower $f$.⁶⁸ This is the precise, measurable statement of the everyday rule that "sharper kernels are noisier."

Putting it together: the tradeoff and the detectability index

Because the same high-frequency weighting drives both the MTF and the NPS, resolution and noise move together under FBP — you cannot raise one without paying in the other. Image scientists fold both into a single detectability index $d^{\prime }$ that predicts how well a defined imaging task (for example, detecting a small low-contrast lesion) can be performed:

$$d^{\prime 2}=\frac{\iint |\text{MTF}(f){|}^2|W_{\text{task}}(f){|}^{2}d{f}_{x}d{f}_y}{\iint \text{NPS}(f)|W_{\text{task}}(f){|}^{2}d{f}_{x}d{f}_y}$$

where $W_{\text{task}}(f)$ encodes the size and contrast of the feature of interest. The key insight is that the optimal kernel is task-dependent: a sharp kernel maximizes $d^{\prime }$ for a high-contrast, high-frequency task (bone trabeculae), while a smooth kernel maximizes $d^{\prime }$ for a low-contrast task (liver lesion). This MTF/TTF–NPS–$d^{\prime }$ framework is the methodology AAPM Task Group 233 standardized for CT performance evaluation, and it is exactly why iterative reconstruction helps: by lowering the NPS without collapsing the MTF, IR raises $d^{\prime }$ at fixed dose, or holds $d^{\prime }$ constant at lower dose.⁶⁷⁸ Specific MTF cutoff frequencies and $\sigma$ values are scanner-, kernel-, and dose-specific and should be measured on your own system.

The Major Siemens Kernel Categories

The table below maps the Siemens kernel families from smooth to sharp against their typical clinical task and the resulting resolution-versus-noise behavior. (Kernel family names and characters below are preserved from Siemens SOMATOM documentation; clinical-task mappings are general guidance.¹)

Body Kernels (B-Family)

Body kernels are used for general body imaging and represent the most common reconstruction choices for chest, abdomen, pelvis, and angiographic studies.

• Br kernels — standard body imaging; balanced sharpness and noise
• Bf kernels — fine-noise body imaging; optimized for a smoother image appearance
• Bl kernels — lung-specific kernels; enhanced edge definition for lung parenchyma
• Bv kernels — vascular kernels; optimized for vessel visualization and CT angiography (CTA)

Head Kernels (H-Family)

Head kernels are optimized for neuroimaging and skull evaluation, with the choice driven by whether soft tissue or bone detail is the primary clinical focus. The brain is a low-contrast task — gray–white matter differences are only a few HU — so head soft-tissue kernels sit toward the smooth end, while bone evaluation calls for the sharp Hc kernels.

• Hr kernels — standard brain imaging
• Hf kernels — fine-noise brain imaging; improved low-contrast detectability
• Hc kernels — crisp kernels; enhanced visualization of bone and high-contrast structures

Quantitative Kernels (Q-Family)

Quantitative kernels are specifically designed to preserve accurate CT-number measurements and minimize reconstruction-related bias, which makes them the correct choice whenever a number, not just an image, is the deliverable.

Common applications include:

• Calcium scoring
• Quantitative lung imaging
• Dual-energy CT analysis
• Radiomics and AI-based quantitative analysis

The most common example is the Qr kernel, which provides consistent CT-number accuracy and improved quantitative reliability. Standardizing on a single quantitative kernel for a given task is what keeps measurements such as Agatston calcium scores reproducible from one study, scanner, or follow-up to the next. The physical reason is that a kernel's frequency response shifts measured CT numbers in small, high-contrast structures — a calcium speck a few pixels across is exactly the kind of high-frequency feature whose apparent attenuation depends on the kernel — so changing kernels mid-program can silently change scores even when nothing else changes.³⁵⁹

Ultra High Resolution Kernels (U-Family)

Ultra High Resolution (UHR) kernels provide maximum spatial resolution and are used for detailed imaging of fine anatomical structures.

Examples include:

• Ur — ultra high resolution regular kernel
• Ub — ultra high resolution body kernel
• Uh — ultra high resolution head kernel
• Ul — ultra high resolution lung kernel

Typical clinical uses include:

• Temporal bone imaging
• Inner ear imaging
• Lung interstitial disease evaluation
• High-resolution bone imaging

Because UHR kernels markedly increase image noise, they are typically used together with iterative reconstruction to keep noise at a diagnostic level. In MTF/NPS terms, a UHR kernel extends the passband to the highest spatial frequencies the detector geometry supports — maximizing $f_{10}$ — but in doing so it raises the high-frequency tail of the NPS so steeply that, under FBP, $\sigma$ would often be diagnostically unacceptable.⁶⁸

Iterative Reconstruction: SAFIRE and ADMIRE

Iterative reconstruction lets you use a sharper kernel without paying the full noise penalty. Modern Siemens CT systems use advanced iterative reconstruction methods such as:

• SAFIRE (Sinogram Affirmed Iterative Reconstruction)
• ADMIRE (Advanced Modeled Iterative Reconstruction)

These techniques reduce image noise while preserving spatial resolution, which allows technologists to select higher-resolution kernels without significantly increasing noise. The practical benefits are:

• Improved image clarity
• Better low-contrast detectability
• Dose optimization, because the noise reduction can be traded for lower tube output

The clinical literature quantifies this. In a prospective chest CT study, SAFIRE allowed an approximately 65% reduction in radiation dose without loss of diagnostic information, with measured objective noise significantly lower than FBP at the same low dose; the authors confirmed that the noise power spectrum retained an FBP-like shape with progressive noise reduction at higher SAFIRE strengths.¹⁰ In head CT, SAFIRE improved contrast-to-noise ratio by roughly 47% versus FBP while keeping objective image sharpness comparable, supporting dose reduction in neuroradiology.¹¹ More general task-based work using model-based and statistical IR has shown dose-reduction potential on the order of tens of percent without compromising modeled detectability — the $d^{\prime }$ framework described above.⁷ SAFIRE and ADMIRE offer selectable strength levels — typically 1 to 5 on current SOMATOM platforms — that let the physicist and technologist tune how aggressively noise is suppressed.

The key takeaway for protocol design is that kernel and reconstruction method are coupled choices: a UHR or high-index kernel that would be unusable with filtered back projection often becomes clinically excellent when paired with ADMIRE or SAFIRE.

Beam Hardening Correction Considerations

Beam Hardening Correction (BHC) reduces artifacts caused by dense structures such as bone or iodinated contrast, where lower-energy photons are preferentially attenuated and bias the reconstructed CT numbers. Important considerations for kernel selection:

• BHC is typically available for kernels with a resolution index ≤45
• BHC is not available for Ultra High Resolution kernels
• BHC improves uniformity and quantitative accuracy in many clinical applications

This is one more reason quantitative tasks tend to live in the lower-index, BHC-eligible kernel range rather than at the UHR end of the spectrum. When dense hardware is in the field of view, kernel choice also interacts with dedicated artifact-reduction tools, as covered in Understanding Metal Artifact Reduction in CT.

Clinical Impact

Kernel selection is not a cosmetic preference — it changes what a radiologist can detect and what a quantitative result reports. Because the kernel sets both the MTF and the NPS, it determines the detectability of every feature in the image, and the right choice is the one that maximizes detectability for the task at hand.⁶⁷⁸ Low-contrast tasks (liver lesions, gray–white differentiation) are noise-limited, so a smooth kernel that lowers $\sigma$ wins; high-contrast, high-frequency tasks (lung interstitium, temporal bone, fracture lines) are resolution-limited, so a sharp kernel that raises the high-frequency MTF wins; and quantitative tasks (calcium scoring, dual-energy, radiomics) are dominated by CT-number accuracy, so a fixed quantitative kernel matters more than maximizing either.³⁵

The cost of getting this wrong is concrete: an over-sharp kernel on a low-contrast study buries lesions in noise, while an over-smooth kernel on a high-resolution study blurs the very detail the exam was ordered to show. Kernel mismatches also undermine comparisons over time, because measured noise, resolution, and CT numbers all shift when the kernel changes.

Practical Tips for Technologists

A short decision guide for routine Siemens CT protocols:

• Use Br kernels for routine chest, abdomen, and pelvis imaging.
• Use Bl or Ul kernels for lung imaging.
• Use Bv kernels for angiographic (CTA) studies.
• Use Qr kernels for calcium scoring and quantitative applications.
• Use UHR kernels only when high spatial resolution is genuinely required, such as temporal bone or interstitial lung disease.
• Pair higher-resolution kernels with ADMIRE or SAFIRE to keep noise in check.
• Standardize kernel selection across like protocols so noise, resolution, and CT-number measurements stay reproducible over time.
• When you change kernel, dose, or iterative strength, treat it as a protocol change: document it and confirm the resulting noise and resolution are still within your physicist's established baselines.

Appropriate kernel selection ensures an optimal balance between image sharpness and noise for each clinical task. It also keeps your image quality measurements stable, which matters every time a medical physicist evaluates the scanner.

Regulatory and Accreditation Considerations

Reconstruction kernel is part of the CT protocol that a qualified medical physicist evaluates during accreditation and annual performance testing. Under the ACR CT Accreditation Program and the ACR-AAPM technical standards, the physicist assesses spatial resolution, image noise, CT-number accuracy, and uniformity, all of which depend directly on the kernel used to reconstruct the phantom and clinical images. Modern performance evaluation increasingly uses the task-based metrics — TTF, NPS, and detectability — formalized in AAPM Task Group 233, and acceptance and constancy testing of CT imaging performance is also addressed by IEC 61223-3-5.⁶⁸

• Reproducibility: Using the same kernel for a given protocol over time keeps annual survey measurements comparable and makes drift easy to detect.
• Quantitative consistency: For calcium scoring and other quantitative tasks, a fixed quantitative kernel is essential for defensible, comparable results.
• State oversight: CT facilities in the states DRPS serves, including Florida (Chapter 64E-5, Florida Administrative Code), Maryland, Virginia, Washington, DC, California, and Nevada, operate under state radiation-control programs that expect documented protocols and qualified-physicist oversight. Diagnostic CT scanners are regulated as radiation-producing machines under FDA and state (Agreement State) radiation-control programs rather than under NRC byproduct-material rules, so kernel and protocol documentation lives within that state framework. Kernel choice is one of the protocol parameters that supports a clean, audit-ready record.

For the broader accreditation picture, see our overview of ACR accreditation physics requirements, and for how kernel choice ties into the dose side of the protocol, see our guide to CTDIvol and DLP dose metrics.

Frequently Asked Questions (FAQs)

How does kernel selection affect CT image quality?

Kernel choice directly controls the tradeoff between sharpness and noise. Smooth (low-index) kernels suppress noise and improve low-contrast detectability, while sharp (high-index) kernels enhance edge detail but increase noise and can alter CT numbers in small or high-contrast structures.

What do the letters and numbers in a Siemens kernel name mean?

A Siemens kernel name has a two-letter marker plus a two-digit resolution index. The first letter is the family (B = body, H = head, Q = quantitative, U = ultra high resolution), the second letter is the subgroup (such as r, f, l, v, c, or p), and the number sets sharpness, with each increment representing roughly a 4% gain in spatial resolution.

What is the difference between filtered back projection and iterative reconstruction?

Filtered back projection applies the kernel to the projection data in a single analytic pass, so resolution and noise are rigidly coupled — a sharper kernel always adds noise. Iterative reconstruction repeatedly refines the image against the measured projections, suppressing noise while largely preserving resolution, so you can run a sharper kernel at the same dose or hold the kernel fixed and lower the dose.

Which Siemens kernel should I use for calcium scoring?

Use a quantitative Qr kernel for calcium scoring. Qr kernels are designed to preserve accurate CT numbers and minimize reconstruction-related bias, which keeps Agatston scores reproducible across scans and scanners.

Do iterative reconstruction methods like SAFIRE and ADMIRE let me use sharper kernels?

Yes. SAFIRE and ADMIRE reduce image noise while preserving spatial resolution, so a higher-resolution kernel can be paired with iterative reconstruction to gain detail without the noise penalty that filtered back projection would impose.

How do physicists measure the resolution and noise a kernel produces?

Spatial resolution is quantified by the modulation (or task) transfer function, and noise by its standard deviation and the noise power spectrum. AAPM Task Group 233 describes these task-based metrics, which a physicist measures on phantom images during commissioning and annual CT testing to verify a kernel behaves as expected.

Why does the kernel matter for ACR CT accreditation and quantitative consistency?

Reconstruction kernel is part of the protocol that drives measured noise, spatial resolution, and CT-number accuracy, all of which a qualified medical physicist evaluates during ACR accreditation and annual surveys. Standardizing kernels across protocols keeps measurements reproducible and audit-ready.

Key Takeaways

• A Siemens kernel name encodes both purpose and sharpness: a two-letter marker (family + subgroup) and a two-digit resolution index, where each increment is roughly a 4% gain in spatial resolution.
• Smooth, low-index kernels reduce noise and aid low-contrast detection; sharp, high-index kernels add edge detail but raise noise and can shift CT numbers.
• The tradeoff is measurable: a sharper kernel raises the high-frequency response of the MTF and simultaneously raises the high-frequency NPS and the noise standard deviation $\sigma$. The detectability index $d^{\prime }$ ties MTF, NPS, and the imaging task together.
• Under filtered back projection, resolution and noise are coupled; iterative reconstruction (SAFIRE, ADMIRE) lowers the NPS while preserving the MTF, enabling sharper or UHR kernels and supporting dose optimization.
• Quantitative (Qr) kernels are the correct choice for calcium scoring and other measurement-driven tasks because they preserve CT-number accuracy.
• Beam Hardening Correction is generally available for kernels with a resolution index ≤45 and is not available for UHR kernels.
• Standardizing kernels across protocols keeps noise, resolution, and CT-number measurements reproducible and accreditation-ready.

How DRPS Can Help

Diagnostic Radiation Physics Services (DRPS) helps imaging facilities translate kernel and reconstruction choices into protocols that are clinically sound, dose-optimized, and accreditation-ready. Our board-certified medical physicists evaluate CT image quality, CT-number accuracy, spatial resolution (MTF/TTF), and noise (magnitude and NPS); support ACR CT accreditation; and review protocols so kernel selection is consistent across scanners and over time. DRPS serves Florida, Maryland, Virginia, Washington, DC, California, and Nevada. To optimize your Siemens SOMATOM protocols or prepare for your next survey, contact DRPS.

Conclusion

Siemens reconstruction kernels give technologists precise control over image resolution and noise, allowing reconstruction to be tailored to each clinical task. Understanding the naming convention and selecting the appropriate kernel improves diagnostic image quality, protects quantitative accuracy, and supports dose optimization. Because the kernel drives the very measurements that physicists evaluate during accreditation — the MTF, the NPS, and the detectability they determine — treating kernel selection as a deliberate, standardized decision is a critical step toward high-quality CT imaging and reliable diagnostic interpretation.

Related Resources

• CT protocol optimization
• Metal artifact reduction in CT
• ACR accreditation physics requirements
• CTDIvol and DLP dose metrics
• Medical physicist consulting

References

1. Siemens Healthineers. SOMATOM CT System Technical Documentation. siemens-healthineers.com
2. Siemens Healthineers. ADMIRE Iterative Reconstruction Technical Guide. siemens-healthineers.com
3. American College of Radiology (ACR). CT Quality Control Manual. accreditationsupport.acr.org
4. Bushberg JT, Seibert JA, Leidholdt EM, Boone JM. The Essential Physics of Medical Imaging, 3rd Edition. Philadelphia: Lippincott Williams & Wilkins; 2012. Open Library
5. Kalender WA. Computed Tomography: Fundamentals, System Technology, Image Quality, Applications. 3rd ed. Erlangen: Publicis Publishing; 2011. WorldCat
6. Samei E, Bakalyar D, Boedeker KL, et al. Performance evaluation of computed tomography systems: Summary of AAPM Task Group 233. Medical Physics. 2019;46(11):e735-e756. doi:10.1002/mp.13763. doi.org
7. Samei E, Richard S. Assessment of the dose reduction potential of a model-based iterative reconstruction algorithm using a task-based performance metrology. Medical Physics. 2015;42(1):314-323. doi:10.1118/1.4903899. doi.org
8. Bhattarai M, Bache S, Abadi E, Samei E. A systematic task-based image quality assessment of photon-counting and energy integrating CT as a function of reconstruction kernel and phantom size. Medical Physics. 2024;51(2):1047-1060. doi:10.1002/mp.16619. doi.org
9. Sato K, Tomita Y, Kageyama R, et al. Method to calculate frequency characteristics of reconstruction filter kernel in X-ray computed tomography. Australasian Physical & Engineering Sciences in Medicine. 2019. doi:10.1007/s13246-019-00819-5. doi.org
10. Kalra MK, Woisetschläger M, Dahlström N, et al. Sinogram-affirmed iterative reconstruction of low-dose chest CT: effect on image quality and radiation dose. AJR American Journal of Roentgenology. 2013;201(2):W235-W244. doi:10.2214/AJR.12.9569. doi.org
11. Korn A, Bender B, Fenchel M, et al. Sinogram affirmed iterative reconstruction in head CT: improvement of objective and subjective image quality with concomitant radiation dose reduction. European Journal of Radiology. 2013;82(9):1431-1435. doi:10.1016/j.ejrad.2013.03.011. doi.org
12. International Electrotechnical Commission. Evaluation and Routine Testing in Medical Imaging Departments — Part 3-5: Acceptance and Constancy Tests — Imaging Performance of Computed Tomography X-Ray Equipment. IEC 61223-3-5. Geneva: IEC. iec.ch