Reconstruction and post-processing

The core idea

Reconstruction creates images; post-processing creates views

The scanner does not acquire a stack of finished coronary pictures. It records x-ray projection measurements over many angles, relates them to the ECG, and mathematically reconstructs voxels with CT attenuation values. A workstation then reformats that image volume into views suited to the clinical question.

The governing rule is simple: every post-processed view must remain accountable to the thin axial data and another orthogonal plane. No single reformat is sufficient for interpreting coronary disease.

Know the boundary

Acquisition choices are locked; reconstruction choices are conditionally flexible

“Flexible” does not mean unlimited. You cannot reconstruct a phase that was never exposed, produce anatomy outside the scan field, restore a missed contrast peak, or recover true high-frequency detail absent from the data. Raw projection data may also be deleted from the scanner after a limited retention period; once only reconstructed DICOM series remain, re-reconstruction options shrink.

Motion is phase-specific

Find the quietest coronary phase—do not accept the default blindly

Reconstruction timing may be expressed as a percentage of the R–R interval or as an absolute number of milliseconds after the R wave. Mid-to-late diastole is often best at a slow, regular heart rate; end-systole is often more reliable as rate rises or diastole shortens. The lecture’s teaching ranges—roughly 70–80% for diastole and 35–45% for systole—are useful search zones, not universal answers.

1. Open the automatically reconstructed phase and inspect the aortic root plus proximal/mid RCA, LAD, and LCx for blurring, duplication, or step artefact.
2. If motion remains and the acquisition window permits, reconstruct nearby phases in small increments rather than guessing one distant phase.
3. Compare coronary-by-coronary: the RCA may move differently from the left system, so a secondary phase can be diagnostic for only one vessel.
4. For irregular R–R intervals, consider absolute-millisecond systolic reconstruction and validated ECG editing where available.

Choose the phase with the sharpest vessel walls and least discontinuity—not simply the phase displaying the largest lumen. Record which phase became the diagnostic series.

How projections are combined

Half-scan and multi-segment reconstruction are not interchangeable

Multi-segment reconstruction is not automatically the cure for irregular rhythm; irregularity is precisely what can make cross-beat combination fail. Use scanner-specific guidance and judge the actual images.

In-plane sampling

Display FOV and matrix determine pixel size—not acquired resolution by themselves

For a square reconstruction, approximate in-plane pixel size is:

A tight cardiac display FOV improves sampling and keeps the heart prominent, but only if the anatomy lies within the acquired scan field. It does not manufacture spatial resolution beyond detector sampling, focal spot, motion, slice thickness, and the reconstruction kernel. An unnecessarily tiny FOV can crop bypass grafts, anomalous vessels, or extracardiac findings.

• Scan FOV: acquisition geometry chosen before the scan; usually cannot be expanded after exposure.
• Display/reconstruction FOV: the portion of raw data mapped into the image matrix; often adjustable before raw data are discarded.
• Matrix: usually 512 × 512, with higher matrices available on selected systems; benefit requires matching acquired resolution and noise.

Through-plane detail

Thin, overlapping slices support near-isotropic coronary analysis

Coronaries are small and oblique to the axial plane. Reconstruct the smallest validated slice thickness the system can support—often approximately 0.5–0.75 mm for conventional CTCA, and thinner on ultra-high-resolution systems—with an overlapping increment. Overlap provides smoother reformats and reduces gaps when following a tortuous vessel.

Keep a thin master dataset for detailed interpretation. Create thicker averaged MPR or MIP slabs from it when a quieter overview is helpful; do not discard the thin data in favour of a single thick series.

Detail versus noise

The reconstruction kernel shapes edges—and the apparent lesion

Kernel names are vendor-specific. Compare kernels on the same phase and geometry, and do not assume the sharpest-looking series is always the most diagnostic. Quantitative plaque analysis should use a validated, consistent protocol.

Noise management

Iterative and deep-learning reconstruction improve the estimate—not the acquisition

Filtered back projection reconstructs the measured projections directly with mathematical filtering. Iterative methods start with an image estimate, model how it would produce projection data, compare that with the measurements, and repeatedly reduce the difference. Modern model-based and deep-learning methods can suppress noise and artefact more effectively.

• Benefit: better noise texture and contrast-to-noise at lower dose or lower kVp; potential improvement in vessel-edge conspicuity.
• Strength setting: higher is not automatically better. Excessive strength can create an unfamiliar plastic or blotchy texture and alter perceived edges.
• Limit: no reconstruction can fully correct a missed contrast bolus, severe cardiac/respiratory motion, cropped anatomy, or inadequate acquired spatial resolution.
• Consistency: reconstruction method and strength affect HU distribution, plaque measurements, calcium blooming, and serial comparison.

Display, not reconstruction

Window width and level change what you see—not the stored HU

Windowing maps a selected HU range to the display grayscale. It can be changed freely on the workstation without reconstructing the image. For 120-kVp CTCA, a practical starting point is around WW 800 / WL 300; the webinar’s approximate WW 700–900 and WL 200–260 reflects the same broad concept. Adjust for actual luminal enhancement, kVp, noise, calcium, and the question.

• Too narrow: contrast and calcium saturate white, blooming increases, and the lumen–plaque boundary may disappear.
• Wider/higher window: can improve inspection of dense calcium or stent struts but may reduce conspicuity of subtle noncalcified plaque.
• Too wide: overall contrast becomes flat and small differences are harder to see.

Windowing can reveal information already present; it cannot reverse partial-volume averaging. Review suspected stenosis at more than one appropriate setting rather than treating one preset as universal.

Choose the right representation

Each post-processing format answers a different question

Diagnostic foundation

Start with axial images; confirm lesions in orthogonal MPR

Thin axial source images preserve grayscale fidelity with the least post-processing distortion. Scroll through the entire dataset to assess coverage, enhancement, noise, motion, step artefact, cardiac structures, and the visible chest. They also provide the reference against which an apparent lesion on any derived view must be checked.

Use oblique MPR to align one plane along the vessel and another exactly perpendicular to the vessel centreline. True short-axis cross-sections are essential for eccentric plaque, lumen area/diameter, lesion length, bifurcations, and stent assessment. If the cross-section is oblique rather than perpendicular, the lumen appears falsely elliptical and stenosis may be misgraded.

Follow the full vessel

Curved MPR is powerful—but only if the centreline is correct

A curved MPR traces a coronary centreline and unfolds the artery into a single longitudinal image. It is excellent for seeing lesion location and the relationship to branches. Rotate the curved view around the vessel axis and review at least two perpendicular longitudinal orientations because an eccentric plaque can be hidden in one projection.

Centreline errors can cut through plaque, leave the lumen, jump into a branch, or straighten across a bend—creating false narrowing or concealing real disease. Correct the line manually and confirm every suspicious segment on orthogonal cross-sections and axial images.

A straightened-vessel MPR presents the traced artery as a line, facilitating rapid lesion search and linked cross-sections. The trade-off is loss of true spatial orientation; use a curved/3D view to understand anatomical relationships.

Projection views

MIP is a map, not a stenosis ruler

Maximum intensity projection selects the highest-attenuation voxel along each viewing ray through a slab. A thin slab—often around 5–10 mm—is useful for displaying longer coronary segments, origins, branches, and bypass grafts with lower apparent noise. Sliding the slab helps track a vessel through the heart.

Because MIP preferentially displays bright contrast and calcium, it can hide low-attenuation plaque behind the opacified lumen, merge adjacent structures, and exaggerate calcified disease. Increasing slab thickness worsens overlap. Never quantify stenosis from MIP alone.

Minimum intensity projection (MinIP) instead displays the lowest-attenuation voxel and can help depict myocardial hypoattenuation or selected airway/valve questions. Like MIP, it is a projection susceptible to overlap and must be correlated with source images.

Three-dimensional communication

Volume rendering is ideal for relationships, not luminal measurement

Volume-rendered techniques assign colour and opacity to the full voxel volume, producing an intuitive 3D roadmap. They are particularly helpful for anomalous origin/course, coronary dominance and gross branching, bypass graft pathways, congenital anatomy, and procedural or surgical planning.

The apparent lumen depends on threshold, opacity, lighting, and segmentation. Calcium, stents, and contrast may merge; noncalcified plaque may be invisible. Cropping can delete anatomy. Therefore VRT—and photorealistic cinematic rendering—must not be used to grade stenosis or to declare a vessel normal.

A repeatable workstation sequence

Build the diagnosis from source data outward

1. Verify the dataset: patient, series, coverage, phase, thickness, kernel, kVp, enhancement, noise, and ECG/motion artefacts.
2. Select the best phase: compare nearby phases and retain a vessel-specific secondary phase if useful.
3. Survey axial images: cardiac and extracardiac structures; locate coronary origins, course, calcification, plaque, stents, and grafts.
4. Trace systematically: LM → LAD/diagonals → LCx/obtuse marginals → RCA/PDA/PL branches; inspect anomalous and graft anatomy as applicable.
5. Use curved/straight MPR: find candidate lesions and verify the centreline.
6. Interrogate each lesion: thin oblique long-axis plus true perpendicular short-axis; compare proximal and distal reference segments.
7. Use MIP/VR selectively: orientation, course, branches, and communication—not standalone grading.
8. Reconstruct problem areas: alternate phase, sharper kernel, thinner slice, different iterative strength, or wider window when data permit.
9. Record limitations: state which segments remain non-diagnostic and why; do not let a normal-looking 3D overview override uncertainty.

High-density targets

Calcium and stents need a deliberate secondary reconstruction

Dense calcium and metallic stent struts spread into adjacent pixels through partial-volume and point-spread effects—blooming—making the lumen look smaller. Start with the best motion-free phase and use a small display FOV, the thinnest appropriate slices, a sharper validated kernel, and suitably wide windowing. On capable systems, high-resolution or spectral reconstructions may help.

These steps reduce but do not abolish blooming. A sharp kernel raises noise; very small stents, thick struts, overlapping stents, heavy calcification, and motion may remain non-diagnostic. Do not infer in-stent patency merely from contrast distal to the stent—collaterals or partial flow can still be present.

Quality control

Common post-processing traps and their fixes

Apply the framework

Worked reconstruction scenarios

Case A · RCA blurred at 75%, LAD and LCx sharp

Do not discard the whole scan or force one global phase. Reconstruct nearby phases available within the exposure window—often searching systolic or adjacent diastolic timing—and retain a vessel-specific RCA series if it is clearly sharper.

Case B · Severe mid-LAD stenosis only on automated curved MPR

Inspect the centreline; it may have crossed an eccentric plaque or left the lumen at a bend. Correct it, rotate the curved view, and confirm with two orthogonal planes plus thin axial images before grading.

Case C · 3.0-mm MIP shows no noncalcified plaque

That does not exclude plaque: MIP shows the highest attenuation and can conceal lower-attenuation wall disease. Review thin axial and cross-sectional MPR images using appropriate windowing.

Case D · Stent lumen unreadable on the routine smooth kernel

Choose the best motion-free phase, reconstruct thinner slices with a small FOV and sharper stent/coronary kernel, adjust windowing, and compare with the routine series. If blooming/noise still prevents lumen assessment, report the segment as non-diagnostic rather than inventing patency.

Case E · Attractive VRT appears to show an anomalous RCA course

Use the VRT as a roadmap, then determine the exact origin and interarterial/intramural relationship on thin axial and oblique MPR images. Threshold-dependent 3D appearances are not sufficient for high-risk anatomy.

Active recall

Self-test: answer before opening each explanation

Printable quick revision

One-page reconstruction recall card

• Raw ≠ axial: projection measurements are reconstructed into viewable source images.
• Phase: diastole often suits slow rates; systole often suits faster/variable rates; inspect the actual vessels.
• Pixel: display FOV ÷ matrix. Smaller pixels do not guarantee better true resolution.
• Slices: thin and overlapping for coronary detail; thicker slabs only as derived overviews.
• Kernel: smooth = quieter; sharp = crisper/noisier. Keep a routine and targeted adjunct series.
• Windowing: display choice only; begin near WW 800/WL 300 at 120 kVp and adapt.
• Curved MPR: full vessel, but centreline error can fabricate disease.
• MIP: map/overview; never standalone stenosis grading.
• VRT: spatial relationships and communication; not lumen quantification.
• Final rule: no single reconstruction is sufficient.

At the workstation, ask: Is this finding present in the patient—or created by phase, plane, centreline, kernel, window, or projection?

Provenance

Sources and scope

This chapter follows the reconstruction/post-processing portions of the locally supplied Webinar 1 talks by Dr Tosha (approximately 25:30–34:50) and Dr Subhajit (approximately 29:05–35:10). Automated captions contain transcription errors; terminology and diagnostic limitations were checked against current professional guidance.

1. SCCT Interpretation and Reporting of Coronary CTA: 2026 Update — current image formats, kernels, windowing, motion mitigation, multi-plane interpretation, calcium and stent pitfalls.
2. SCCT Guidelines for the Performance and Acquisition of Coronary CTA (2016) — phase timing, reconstruction thickness, kernels, FOV, and acquisition-window limits.
3. SCCT Standards for Quantitative Assessments by CCTA — effects of contrast, kVp, kernel, spatial resolution, and protocol consistency on quantitative plaque assessment.
4. Multisociety Expert Consensus on Coronary CTA — source-image review, orthogonal reformats, MIP, curved MPR, and post-processing artefacts.

Educational material only. Vendor kernel names, phase tools, iterative/deep-learning strengths, and raw-data retention vary. Apply the validated protocol for the scanner and keep uncertain segments explicitly non-diagnostic. External references require internet; this chapter and self-test work offline.