Left/Right Mirror Difference Map
CT paranasal sinuses · Series3 (bone kernel) · scan of 2026-08-08 · the head is reflected about its own best-fit mid-sagittal plane and subtracted from itself, so any left/right difference shows up as colour.
Read this before reading the colours. Mild facial asymmetry and some degree of septal deviation are near-universal and usually of no consequence. Human eyes are poor at spotting asymmetry and excellent at spotting colour — that is precisely why this map exists, and precisely why it invites over-reading. Nothing on this page is a severity grade, a normal range, or a verdict. Questions about what any of it means belong to an ENT / radiologist with the actual scan.
1 · Did the plane fit actually work?
Everything below depends on one thing: the fitted mid-sagittal plane. A bad plane fit produces a beautiful, symmetric-looking asymmetry map that is entirely wrong. So check it by eye first: the magenta line should run through the crista galli, the septum and the vomer.
Fitted by maximising normalised cross-correlation between the head and
its own mirror image (head-masked, multi-resolution search):
yaw -2.08°, roll -1.86°,
offset -1.66 mm from patient x = 0; NCC = 0.6836.
2 · Slice viewer — original vs mirrored
Coronal slices (the meaningful view for the septum and sinuses). Drag across the image to wipe between the original head and its mirror image: where the two agree the wipe line is invisible; where they differ, structures jump. Viewer right = patient LEFT (radiological convention).
64 coronal slices, ~2.4 mm apart, nose tip → sphenoid. Arrow keys change slice.
3 · Difference maps
Signed difference in HU between each voxel and its mirror twin, drawn over the anatomy. Red: this side is denser than the mirrored opposite side; blue: less dense. Faint colour is mostly interpolation and posture noise — the colour ramp is deliberately transparent below ~100 HU.
4 · Nasal cavity close-up
This close-up picks up both a deviated septum and any compensatory enlargement of the turbinate on the opposite side. Those are two different things — one is bone/cartilage position, the other is mucosal soft tissue — and a difference map cannot separate bone from mucosa. On a bone-kernel series, mucosal detail is additionally unreliable.
5 · Numbers (made-up, no reference range)
Global asymmetry index = mean |HU difference| over all head voxels: 146.7 HU; 12.30% of head voxels differ from their mirror twin by more than 300 HU. This index was invented for this page. It has no reference range, no units that mean anything clinically, and no diagnostic value. It mixes bone, air and soft tissue, and it rises with posture tilt, dental work and interpolation error just as happily as with anatomy.
| region (approximate z-band) | mean |ΔHU| | voxels >300 HU |
|---|---|---|
| dental/mandible band (~lower 25% of head) | 158.3 | 11.98% |
| palate–maxillary band (~25–50%) | 169.1 | 15.40% |
| orbit–ethmoid band (~50–75%) | 173.0 | 15.21% |
| frontal/calvarial band (~upper 25%) | 90.1 | 6.71% |
| nasal cavity box (crude percentile bounds of near-midline air) | 304.7 | 32.08% |
6 · How much of this survives a plane-fit error?
The fitted plane was deliberately rotated by ±2° in yaw and roll (without refitting) and the whole map recomputed; +0.5° rows are included for scale, since the fit itself converged to within ~0.05° across four resolutions. If the map changed completely, the colours would be an artefact of the fit; if it barely changed, they reflect the anatomy.
| perturbation | correlation with original map | Dice of |Δ|>300 voxels | original flags still flagged | asymmetry index change |
|---|---|---|---|---|
| yaw+2° | 0.549 | 0.605 | 66.3% | +23.0% |
| yaw-2° | 0.601 | 0.611 | 67.0% | +27.4% |
| roll+2° | 0.532 | 0.560 | 69.5% | +60.9% |
| roll-2° | 0.506 | 0.553 | 66.5% | +52.9% |
| yaw+0.5° | 0.913 | 0.815 | 82.8% | +2.4% |
| roll+0.5° | 0.865 | 0.772 | 80.6% | +9.5% |
Rule of thumb for reading the maps: coarse blobs that persist under these perturbations are probably real left/right differences; thin rims hugging bone edges are exactly what a small plane error (or normal skull curvature) produces, and should not be over-read.