What Is a Reconstruction Kernel? Soft vs Sharp Filter

We saw the difference between a CT's "bone window" and "lung window" in the window/level article — but that only changes the display. A deeper difference is set while the image is still being formed: the kernel (filter) chosen during reconstruction. From the same raw data, you can produce a smooth, clean image or a sharp but noisy one. This choice sets the balance between resolution and noise.

What is a kernel?

In CT the image is built by back-projecting the raw projection data; at the heart of that process is a convolution kernel (reconstruction filter).¹ The kernel sets how the data is filtered — which spatial frequencies are emphasized and which suppressed. The key point: the kernel changes not the raw data but how it is processed; so multiple image series can be produced from the same scan with different kernels.¹

Resolution–noise

The kernel choice is a trade-off between the two faces of image quality:

Soft (soft-tissue) kernel: suppresses high frequencies → low noise, but low spatial resolution. Ideal for low-contrast soft tissues like liver or brain, where noise is the real enemy.
Sharp (bone) kernel: emphasizes high frequencies → high spatial resolution, but high noise. Ideal for high-contrast fine structures like bone trabeculae or lung parenchyma, where noise is tolerable but fine detail is essential.¹

The same edge is rounded and cleaned by a soft kernel (low noise, low resolution); a sharp kernel preserves the edge but raises noise. There is no single "best" kernel — it depends on the task.¹

Which for which task?

The right kernel depends on the clinical question. For soft-tissue assessment (e.g. a liver lesion), a soft kernel that suppresses noise; for bone or lung assessment, a sharp kernel that preserves fine detail. Often the same scan is reconstructed as multiple series with different kernels — e.g. a chest CT with soft-tissue, lung and bone kernels. Because the kernel choice directly sets MTF (resolution) and noise, it is the adjustable heart of image quality. Iterative/deep-learning reconstruction (FBP, IR, DLR) stretches this balance further.

In a nutshell

Kernel = the filter chosen at reconstruction; it sets the resolution–noise balance. Soft kernel = clean but smooth (soft tissue); sharp/bone kernel = crisp but noisy (bone/lung). Many series can come from the same data. No single "best" — it depends on the task.

Reconstruction (FBP, IR, DLR) · What Is MTF? · DICOM and Window/Level · Image Quality

References

Bushberg JT, Seibert JA, Leidholdt EM, Boone JM. The Essential Physics of Medical Imaging, 3rd ed. Lippincott Williams & Wilkins, 2011. §4 (convolution kernel) ve §10 (CT): geri-projeksiyonda kullanılan rekonstrüksiyon filtresi/kerneli görüntüyü belirler; aynı veriden yumuşak doku filtresi (düşük gürültü, düşük çözünürlük) ve kemik filtresi (yüksek çözünürlük, yüksek gürültü) farklı sonuç verir (ACR çözünürlük modülü, yumuşak doku vs kemik filtresi). Sayfa numaraları bu baskıya aittir.
İlişkili: Rekonstrüksiyon (FBP, IR, DLR) · MTF Nedir? · Görüntü Kalitesi

Note: This content is for education; for clinical decisions or regulatory compliance, consult a qualified medical physicist and current regulations.

← All articles