Dev

CBCT — Dev Tier

(20 scenes)

Blind evaluation: 20 procedural 256³ phantoms (anatomy-inspired, based on CQ500/AAPM/CBCTLiTS/MMDental characteristics).

What you get

Cone-beam projections (y), ideal geometry (H), and spec ranges. No ground truth.

How to use

Apply your pipeline from Public tier. Self-check via consistency metric. Ground truth scored server-side.

What to submit

Reconstructed volumes and corrected mismatch spec. Scored server-side.

Parameter Specifications

🔒

True spec hidden — estimate parameters from spec ranges below.

Parameter Spec Range Unit
source_offset_x -1.5 – 2.5 mm
source_offset_z -1.2 – 1.8 mm
detector_tilt -0.4 – 0.6 deg
detector_shift_u -2.2 – 3.8 px
beam_hardening -0.035 – 0.115
scatter_fraction -0.02 – 0.08

Dev Tier Leaderboard

# Method Score PSNR SSIM Consistency Trust Source
1 CTFormer + gradient 0.782 33.99 0.961 0.79 ✓ Certified Wang et al., MICCAI 2023
2 DuDoTrans + gradient 0.772 33.21 0.954 0.79 ✓ Certified Wang et al., IEEE TMI 2022
3 DiffusionCBCT + gradient 0.728 29.32 0.906 0.86 ✓ Certified Gao et al., Med. Phys. 2024
4 Learned Primal-Dual + gradient 0.719 28.48 0.891 0.89 ✓ Certified Adler & Oktem, IEEE TMI 2018
5 DuDoNet + gradient 0.700 28.55 0.892 0.79 ✓ Certified Lin et al., CVPR 2019
6 Metal-AR-Net + gradient 0.669 26.5 0.846 0.83 ✓ Certified Zhang & Yu, IEEE TMI 2018
7 FDK + gradient 0.665 26.38 0.842 0.82 ✓ Certified Feldkamp et al., J. Opt. Soc. Am. A 1984
8 FBPConvNet + gradient 0.647 25.36 0.813 0.84 ✓ Certified Jin et al., IEEE TIP 2017
9 TV-ADMM + gradient 0.628 24.26 0.778 0.87 ✓ Certified Boyd et al., Found. Trends 2011

Visible Data Fields

y H_ideal spec_ranges

Dataset

Format: HDF5
Scenes: 20

Scoring Formula

0.4 × PSNR_norm + 0.4 × SSIM + 0.2 × (1 − ‖y − Ĥx̂‖/‖y‖)

PSNR: 40% SSIM: 40% Consistency: 20%
Back to CBCT