Partial Volume Effect

PVE occurs when a voxel straddles the boundary between two or more tissue types with different signal intensities. The resulting voxel value is a weighted average of the constituent tissues' signals, proportional to their fractional volume occupancy. For example, a voxel containing 60% gray matter and 40% white matter will display an intermediate intensity that represents neither tissue accurately. This averaging obscures fine anatomical details and creates the illusion of a gradual transition where a sharp boundary actually exists.

The severity of PVE is directly proportional to slice thickness. Thicker slices produce larger voxels that are more likely to encompass multiple tissue types, exacerbating the effect. Key relationships:

• Thin slices (≤1mm): Minimize PVE but reduce signal-to-noise ratio (SNR)
• Thick slices (≥5mm): Improve SNR but significantly increase PVE
• Isotropic voxels: Equal dimensions in all axes provide the best trade-off Modern high-resolution CT and MRI protocols use sub-millimeter slice thicknesses specifically to mitigate PVE in critical applications like angiography and cortical surface mapping.

PVE introduces systematic errors in volumetric measurements and tissue classification. When segmenting small structures, PVE causes:

• Volume underestimation: Thin structures may fall below partial volume threshold
• Boundary displacement: Apparent edges shift from true anatomical boundaries
• Intensity contamination: Adjacent high-intensity structures artificially elevate values in neighboring regions In PET/CT quantification, PVE leads to underestimation of tracer uptake in small lesions, potentially causing false-negative diagnoses. Correction techniques like geometric transfer matrix (GTM) and Müller-Gärtner methods are essential for accurate SUV measurements.

At tissue interfaces, PVE creates a gradient of intermediate intensities that smooths what should be a discrete boundary. This is particularly problematic in:

• Cortical surface extraction: Gray-white matter boundaries become ambiguous
• Tumor margin delineation: Infiltrative edges are indistinguishable from PVE blurring
• Vessel wall imaging: Small vessel lumens may appear partially occluded The effect is most pronounced when the in-plane resolution is coarse relative to the structure size. Structures smaller than 2-3 times the voxel dimension are particularly susceptible to complete signal loss or mischaracterization.

Several algorithmic approaches exist to mitigate PVE:

• Voxel-based correction: Uses high-resolution anatomical priors to model tissue fractions within each voxel
• Region-based correction: Applies geometric models assuming known structure shapes
• Deconvolution methods: Attempt to recover the true signal distribution by modeling the point spread function (PSF)
• Deep learning approaches: CNNs trained on paired low/high-resolution data learn to super-resolve partial volume artifacts Modern iterative reconstruction algorithms in CT inherently model the finite voxel size, providing some PVE compensation during image formation rather than as a post-processing step.

PVE is fundamentally governed by the imaging system's point spread function (PSF), which describes how a point source is blurred in the acquired image. The PSF's full width at half maximum (FWHM) determines the minimum resolvable distance between two structures. When the PSF extends across tissue boundaries, signal from one region spills over into adjacent voxels. In PET imaging, the PSF is typically 4-7mm FWHM, making PVE the dominant source of quantification error for lesions smaller than 2-3 times this value. PSF modeling during reconstruction is now standard in clinical PET/CT systems to reduce this effect.