Public

Diffusion MRI — Public Tier

(3 scenes)

Full-access development tier with all data visible.

Measurements (y), ideal forward operator (H), spec ranges, ground truth (x_true), and true mismatch spec.

Load HDF5 → compare reconstruction vs x_true → check consistency → iterate.

Reconstructed signals (x_hat) and corrected spec as HDF5.

✓

True spec visible — use these exact values for Scenario III oracle reconstruction.

Parameter	Spec Range	True Value	Unit
b_value_error	-3.0 – 6.0	1.5	%
eddy_current	-0.5 – 1.0	0.25	voxels
gradient_direction	-1.0 – 2.0	0.5	deg
susceptibility	-1.0 – 2.0	0.5	voxels

Method: CPU_baseline — Mismatch parameter: nominal

17.61 dB

SSIM 0.3262

13.04 dB

SSIM 0.0468

16.34 dB

SSIM 0.1224

Per-scene breakdown (4 scenes)

Scene	PSNR I	SSIM I	PSNR II	SSIM II	PSNR III	SSIM III
scene_00	16.58	0.3251	12.51	0.0447	15.93	0.1214
scene_01	18.88	0.3252	13.74	0.0480	17.02	0.1162
scene_02	18.30	0.3298	13.37	0.0496	16.55	0.1259
scene_03	16.68	0.3248	12.53	0.0450	15.85	0.1262

#	Method	Score	PSNR	SSIM	Consistency	Trust	Source
1	DiffusionDTI + gradient	0.858	37.87	0.982	0.93	✓ Certified	Gao et al., NeurIPS 2024
2	PhysDiffMRI + gradient	0.818	35.69	0.972	0.86	✓ Certified	Chen et al., MRM 2024
3	SwinDTI + gradient	0.802	34.43	0.964	0.86	✓ Certified	Wang et al., MICCAI 2023
4	DTIFormer + gradient	0.783	32.87	0.951	0.87	✓ Certified	Liu et al., MICCAI 2022
5	DWIML-Net + gradient	0.764	30.61	0.926	0.94	✓ Certified	Qin et al., IEEE TMI 2019
6	DnCNN-DTI + gradient	0.693	27.18	0.862	0.88	✓ Certified	Golkov et al., IEEE TMI 2016
7	CHARMED + gradient	0.668	25.48	0.817	0.93	✓ Certified	Assaf & Basser, NeuroImage 2005
8	SHORE + gradient	0.614	23.16	0.737	0.94	✓ Certified	Merlet & Deriche, MRM 2013
9	DTI-FIT + gradient	0.512	19.76	0.587	0.9	✓ Certified	Behrens et al., MRM 2003

y H_ideal spec_ranges x_true true_spec

Format: HDF5

Scenes: 3

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

PSNR: 40% SSIM: 40% Consistency: 20%