Calibrated MD intervals gave 95% minimum coverage in fixed ds008695 tensor designs under bounded log curvature and known Gaussian noise

We tested diffusion MRI scan designs in OpenNeuro ds008695 version 1.0.0. We derived a sharp mean diffusivity (MD) bias bound after linear interpolation of b=0 log references. The model used a fixed tensor and bounded baseline second differences by original volume index. We tested ordinary least squares (OLS) and generalized least squares (GLS). Raw log errors were independent Gaussian values with known variance. The main design had 292 diffusion volumes and six references. The fixed curvature scale limited local log departures to log(1.01). It did not limit total scan drift. At raw log-noise standard deviation 0.01, usual 95% intervals had minimum coverage of 72.9228% for OLS and 68.7293% for GLS. Both intervals included shared reference noise. Calibrated intervals gave 95% minimum coverage over the allowed baseline class. Their coverage in 100,000 raw noise draws was 95.033% and 95.053%. Independent linear programs confirmed six sharp bounds. A fixed subset near references reduced the GLS bias bound by 76.9992% and increased noise standard deviation by 12.5414%. The package includes the proof, audit tool, source files, provenance, attaining baselines, and all fixed checks. All 136 source files passed a fresh download hash check. All 36 result files were reproduced byte for byte from those downloaded files. These are scan design and model results. No measured image signal, scanner drift, or clinical effect was tested.

diffusion MRI · mean diffusivity · signal drift · reference interpolation · bounded curvature · coverage · generalized least squares · scan design