Multi-Planar Reconstruction (MPR)

MPR quality is directly dependent on isotropic voxel acquisition—where the x, y, and z dimensions of each voxel are equal. Modern multi-detector CT scanners acquire sub-millimeter slices (e.g., 0.625 mm) to create cubic voxels, enabling reconstructed coronal and sagittal planes to match the spatial resolution of the original axial plane. Without near-isotropic source data, reformatted images suffer from stair-step artifacts and degraded z-axis resolution.

MPR algorithms extract a 2D slice by sampling voxel intensities along a user-defined plane through the 3D volume. Standard orthogonal planes include:

• Coronal: Divides body into anterior/posterior sections
• Sagittal: Divides body into left/right sections
• Axial (transverse): The native acquisition plane
• Oblique: Any non-orthogonal angle, critical for structures like the aortic arch or pancreatic duct
• Curved planar: Follows a tortuous anatomical path, such as a coronary artery, projected onto a single 2D image

When the resampling plane passes between discrete voxel centers, the MPR engine must estimate intensity values. Common interpolation strategies include:

• Nearest neighbor: Fastest, assigns the value of the closest voxel; produces blocky artifacts
• Trilinear interpolation: Weighted average of the 8 surrounding voxels; balances speed and smoothness
• Bicubic/spline interpolation: Higher-order curves using a larger neighborhood of voxels; produces the smoothest results at higher computational cost
• Lanczos resampling: Uses a windowed sinc function; preserves high-frequency detail but may introduce ringing artifacts at sharp edges

Slab MPR (also called thick-slice MPR) averages a user-defined thickness of voxels along the viewing ray rather than sampling a single infinitesimal plane. This technique improves signal-to-noise ratio and provides anatomical context. Common applications include:

• Average Intensity Projection (AIP): Mean voxel value across the slab; mimics a thicker physical slice
• Minimum Intensity Projection (MinIP): Displays the lowest attenuation voxel; excellent for visualizing the tracheobronchial tree and lung parenchyma
• Slab thickness typically ranges from 3 mm to 15 mm depending on the anatomical region and diagnostic requirement

Modern PACS workstations and advanced visualization servers perform MPR in real time, allowing radiologists to dynamically scroll, rotate, and adjust plane orientation with sub-second latency. This is achieved through:

• Pre-cached volume data loaded into GPU memory
• Hardware-accelerated texture mapping using 3D texture units on modern GPUs
• Adaptive level-of-detail rendering that reduces interpolation complexity during rapid interaction
• Typical clinical workflows involve synchronized triple-plane views (axial, coronal, sagittal) with a 3D volume-rendered reference image for spatial orientation

MPR eliminates the need for additional patient scanning and radiation exposure while providing diagnostic planes that are physically impossible to acquire directly. Key benefits include:

• Fracture assessment: Sagittal and coronal reformats of the spine reveal vertebral compression fractures invisible on axial slices alone
• Vascular analysis: Curved planar reformations unwrap tortuous vessels like the aorta into a single longitudinal view for stenosis measurement
• Oncological staging: Multi-planar views improve tumor margin delineation and relationship to adjacent structures
• Pre-surgical planning: Oblique planes aligned with surgical approaches provide intuitive anatomical roadmaps for orthopedic and neurosurgical procedures

The fundamental unit of a 3D volumetric dataset, analogous to a pixel in 2D. Each voxel represents a sampled value—typically a Hounsfield Unit (HU) in CT or signal intensity in MRI—on a regular 3D grid. MPR algorithms resample these voxels along arbitrary planes to generate the new 2D views.

• Isotropic voxels (equal dimensions in all axes) produce the highest quality MPR images with no directional distortion.
• Anisotropic voxels (e.g., 0.5mm x 0.5mm x 3.0mm) cause stair-step artifacts in reformatted planes not parallel to the acquisition.

The mathematical process of estimating voxel intensity values at non-integer grid positions during MPR generation. When a reformatted plane cuts diagonally through the original volume, the exact pixel locations on that plane rarely align perfectly with the original voxel centers.

• Nearest Neighbor: Fastest, preserves original HU values exactly, but produces blocky results.
• Trilinear Interpolation: Computes a weighted average of the 8 surrounding voxels, producing smooth results suitable for diagnostic viewing.
• Bicubic/Spline Interpolation: Higher-order methods used for maximum quality in 3D rendering, at increased computational cost.

A sequence of DICOM images acquired in a single scan that share identical modality, orientation, and temporal parameters. This ordered stack of 2D slices forms the volumetric dataset from which MPR is computed.

• The DICOM header for each slice contains spatial metadata (Image Position Patient, Image Orientation Patient, Pixel Spacing) that the MPR engine uses to construct a geometrically accurate 3D volume.
• A single CT chest scan typically contains 300–700 axial slices, each a 512×512 matrix, forming the input for coronal and sagittal reformations.

The physical depth of each acquired cross-sectional image plane, directly governing the spatial resolution of MPR outputs. Thinner slices yield isotropic datasets that produce reformatted views with equivalent resolution in all planes.

• Modern multi-detector CT scanners routinely acquire at 0.5–0.625mm slice thickness, enabling diagnostic-quality coronal and sagittal MPR.
• Thick slices (>3mm) introduce partial volume effects that blur boundaries and create stair-step artifacts in reformatted planes, limiting diagnostic utility.

A volume rendering technique often used alongside MPR for vascular assessment. While MPR displays a single thin-slice plane through the volume, MIP projects the brightest voxel along each viewing ray onto a 2D image.

• MIP is the standard method for visualizing contrast-enhanced vessels in CT angiography, as it highlights opacified lumens while suppressing lower-attenuation soft tissue.
• Sliding thin-slab MIP combines both techniques: a small range of MPR slices are MIP-projected to improve vessel continuity without obscuring overlapping structures.

An advanced extension of MPR where the reformatted plane follows a user-defined curved path through the volume—typically along the centerline of a tortuous anatomical structure like a coronary artery or colon.

• CPR straightens the vessel of interest, allowing the entire length to be visualized in a single 2D image for stenosis assessment.
• Requires a prior segmentation step to extract the centerline, often performed by a deep learning model such as a 3D U-Net.