Micro-CT Image Quality Toolbox

MTF, NPS, NEQ, TTF and detectability index (d') for reconstructed CT volumes, following AAPM Report 233 and ISO 12233.

Volumes are (z, y, x) arrays from .npy, .npz or GE .vff files. Regions are given in millimetres, so one ROI table works at any voxel size.

Install

pip install micro_ct_image_quality_toolbox

Python 3.10+. Import name microct_iq_toolbox. Add [photutils] for exact circle/pixel overlap in the radial averages (the NumPy default agrees to <1 %).

Usage

ct-demo --out demo/                                        # synthetic phantom, full pipeline, ~10 s

ct-find-rois --template seed.json                          # edit the rough positions
ct-find-rois reference.vff --seed seed.json --out rois.json
ct-report volume.vff --rois rois.json --out results/

ct-find-rois measures the edge angle, insert centres and noise ROIs once per phantom and writes rois.json with a verification figure. ct-report runs every metric on a volume and writes the curves (.npz + .json), metrics.json and the figure above. Voxel size is read from the VFF header or a .json sidecar; use --pixel-size for a bare .npy.

Single metrics with pixel regions:

ct-mtf --input edge.npy --crop_indices 270,664,522,640 --pixel_size 0.075 --edge_angle 5.4 --output_dir r/
ct-nps --input flat.npy --roi_bounds "178,294,510,626;258,374,750,866" --pixel_size 0.075 --output_dir r/
ct-neq --mtf r/mtf.npz --nps r/nps.npz --output_dir r/
ct-dprime npw --mtf r/mtf.npz --nps r/nps.npz --contrast 100 --output_dir r/
ct-ttf --input inserts.npy --centre_pixels "1335,2141;765,1914" --radius 120 --pixel_size 0.075
ct-metrics --config config.json

From Python:

import microct_iq_toolbox as iq

r = iq.mtf(vol, [270, 664, 522, 640], pixel_size=0.075, edge_angle=5.4)
r.x, r.y, r.summary["mtf50"]

rois = iq.PhantomROIs.load("rois.json")
frame = iq.Frame.for_image(vol, pixel_size=0.075)
metrics = iq.run_metrics(vol, frame, rois, out_dir="results")

Each function returns a MetricResult with the curve, a summary dict and save(). The raw calculators (iq.get_MTF, ...) are still available.

Notes:

• The edge angle is an input; a wrong value biases the MTF.
• TTF inserts below CNR 5 are reported but excluded from the average.
• The detectable disc size is where d' crosses 3 (Rose). It is nan if the curve never gets there.
• Regions that do not fit the volume are listed under skipped, never clamped.

Development

pip install -e ".[dev]"
pytest

License

MIT