Hounsfield Unit (HU)

The Hounsfield Unit (HU) is a linear transformation of the measured linear attenuation coefficient (μ). The scale is calibrated such that distilled water at standard temperature and pressure is exactly 0 HU, and air is exactly -1000 HU. The formula is: HU = 1000 × (μ_tissue - μ_water) / (μ_water - μ_air). This normalization ensures that scanners from different vendors produce comparable quantitative values, enabling consistent tissue characterization across devices.

Different biological tissues exhibit distinct, predictable HU ranges due to their physical density and atomic number. Fat typically measures between -100 and -50 HU due to its low density. Soft tissues like muscle, liver, and kidney fall between +20 and +70 HU. Acute hemorrhage appears hyperdense at +60 to +90 HU. Calcified bone and cortical bone exceed +400 HU, often reaching +1000 HU or more. Metal implants can exceed +3000 HU, causing severe beam-hardening artifacts.

The Hounsfield scale is designed to be linear with respect to the attenuation coefficient. However, the Partial Volume Effect introduces non-linear artifacts at boundaries. When a single voxel contains a mixture of two tissues (e.g., bone and soft tissue), the resulting HU value is a weighted average of the two, not a distinct intermediate tissue. This blurs boundaries and can misrepresent small lesions. High-resolution, thin-slice acquisitions minimize this effect by reducing voxel dimensions.

Intravenous iodinated contrast agents dramatically increase the radiodensity of vascular structures and perfused tissues. Unenhanced blood measures +30 to +45 HU, while contrast-enhanced blood can exceed +100 to +300 HU depending on the injection protocol and scan timing. This transient hyperdensity is critical for CT angiography and lesion characterization, allowing radiologists to distinguish patent vessels from soft tissue and to assess organ perfusion dynamics.

The human eye can distinguish only ~30 shades of gray, but a CT image contains 2000+ HU values. Windowing maps a specific range of HU values to the full grayscale display. The Window Width (WW) defines the range of HU values displayed, and the Window Level (WL) defines the center. For example, a lung window (WL -600, WW 1500) visualizes the pulmonary parenchyma, while a bone window (WL +400, WW 1800) is required to see osseous detail without saturation.

Several physical phenomena corrupt the accuracy of HU measurements. Beam hardening causes cupping artifacts and dark streaks, artificially lowering HU values in the center of an object. Metal artifacts from implants cause severe streaking and photon starvation, rendering HU values unreliable. Motion artifacts from patient movement create misregistration and blurring. Quantum mottle (noise) introduces statistical variation, reducing the precision of HU measurements in low-dose scans.

The process of mapping Hounsfield Unit values to grayscale display values. The window level sets the center HU value, while the window width defines the range. This optimizes contrast for specific tissues:

• Lung window: W=1500, L=-600 (visualizes airspaces)
• Bone window: W=2000, L=500 (visualizes osseous detail)
• Soft tissue window: W=400, L=40 (visualizes organs) Without proper windowing, subtle pathologies with narrow HU ranges become invisible.

A volumetric pixel representing a value on a regular grid in 3D space, serving as the fundamental unit for CT and MRI image reconstruction. Each voxel stores a Hounsfield Unit in CT imaging, encoding the average radiodensity of the tissue within that volume. The voxel's dimensions are determined by the pixel spacing and slice thickness, directly influencing spatial resolution and the partial volume effect.

An imaging artifact where a single voxel contains a mixture of multiple tissue types, resulting in a blurred, averaged Hounsfield Unit value that misrepresents both tissues. This degrades boundary definition and can obscure small lesions. Mitigation strategies include:

• Acquiring thinner slice thickness
• Using higher-resolution reconstruction matrices
• Applying partial volume correction algorithms during segmentation

A binary or multi-class label map that classifies each voxel in a volumetric image as belonging to a specific anatomical structure or lesion. Hounsfield Unit thresholds often provide the initial seed for segmentation algorithms:

• Bone: > 300 HU
• Soft tissue: 20–100 HU
• Fat: -100 to -50 HU
• Air: -1000 to -900 HU Deep learning models like 3D U-Net refine these initial HU-based masks.

An analytic reconstruction algorithm that applies a high-pass filter (ramp filter) to raw projection data before back-projecting it across the image grid. This process converts attenuation measurements into Hounsfield Unit values. While computationally fast, FBP amplifies noise and produces streak artifacts in low-dose scans. Modern systems increasingly replace FBP with Iterative Reconstruction (IR) or Deep Learning Reconstruction (DLR) for superior noise-to-HU fidelity.

A class of algorithms designed to mitigate severe streaking and dark-band artifacts caused by metallic implants (e.g., hip prostheses, dental fillings). These artifacts corrupt the true Hounsfield Unit values, rendering adjacent anatomy non-diagnostic. MAR techniques include:

• Projection interpolation: Replacing corrupted sinogram data
• Normalized MAR: Normalizing projections before interpolation
• Deep learning MAR: Using CNNs to restore true HU values Accurate HU restoration is critical for radiation therapy planning.