eXplore CT 120 Reconstruction

Cone-beam CT reconstruction for the GE eXplore CT 120 micro-CT scanner. Supports FDK (analytic), ASTRA (SIRT, CGLS), and TIGRE (OS-SART) backends.

Pipeline

Raw projections → flat-field + log → BHC → ring correction
  → FDK: cone-weight + ramp filter + Parker + backprojection
    OR iterative: ASTRA SIRT / TIGRE OS-SART
  → [bone BHC (FDK only): segment → forward-project → re-reconstruct]
  → physics HU → two-point calibration (air→-1000, water→0) → VFF

BHC (water, 80 kVp) and ring correction are on by default. HU calibration measures air and water/tissue directly from the reconstructed volume (standard CT two-point formula) — self-calibrating regardless of filter or BHC settings.

Usage

# Standard FDK (BHC + ring correction + two-point HU are all defaults)
python -m reconstruction.run_fdk_recon data/scans/Scan_1988 \
    --fov-xy 93.5 --fov-z 70

# Add bone BHC (Joseph & Spital two-pass)
python -m reconstruction.run_fdk_recon data/scans/Scan_1988 \
    --bone-bhc --fov-xy 93.5 --fov-z 70

# ROI reconstruction (mouse lung)
python -m reconstruction.run_fdk_recon data/scans/Scan_1510 --roi auto

# Iterative (ASTRA SIRT)
python -m reconstruction.run_iterative_recon data/scans/Scan_1988 \
    --backend astra --algorithm SIRT3D_CUDA --iterations 100

Run --help for full argument lists.

Key Flags

Flag	Default	Description
--bhc-coeffs c1 c2	0.856 0.21	Sinogram-domain water BHC polynomial (80 kVp)
--no-bhc		Disable BHC
--bone-bhc	off	Two-pass bone BHC (Joseph & Spital, FDK only)
--bone-bhc-threshold	1500	HU threshold for bone segmentation
--bone-bhc-hu	3100	Monochromatic bone HU (from scan.xml BoneHU)
--ring-correction	on	Sinogram-space ring artifact correction
--roi auto	off	ROI from SubVolumeCoordinates.xml

Installation

pip install -e .                # Core (FDK, requires PyTorch)
pip install astra-toolbox       # Optional: ASTRA iterative

Scanner Specifics

Tailored for the GE eXplore CT 120: cone-beam geometry, VFF projections, scan.xml metadata, bright.vff/dark.vff flat-field. Algorithms are general-purpose — adapting to other scanners requires only changing geometry and file I/O.