Patient-level derived dataset: Simulated Post-Operative Liver Resection Registration Benchmark

Description: This repository contains patient-level derived data and analysis code from a study evaluating deformable image registration (ANTs SyN and Elastix B-spline) under simulated partial-overlap liver resection conditions, using 3D MRI from the publicly available Duke Liver Dataset (Macdonald et al., 2023). It accompanies the manuscript "Simulated Post-Operative Liver Resection in 3D MRI for FLR Estimation and Cavity-Masked Registration Robustness Evaluation," submitted to PLOS ONE (manuscript ID PONE-D-25-68351). Dataset For each of 10 patients, 2 simulated resection sides (left/right lobectomy), and 2 registration methods, the dataset reports: total and preserved liver volume, future liver remnant percentage (FLR%), the corrected identity-transform Dice baseline, registered Dice similarity coefficient, registered 95th-percentile Hausdorff distance (HD95), and full distributional statistics (mean, standard deviation, median, 5th/95th percentiles, minimum, maximum, and percentage of voxels outside the range [0.8, 1.25]) for both Jacobian determinant and displacement magnitude within the future liver remnant mask. Each patient identifier is mapped to its corresponding case identifier in the original Duke Liver Dataset, allowing the ten cases used in this study to be identified unambiguously. Two displacement magnitude values (patient P10, left and right lobectomy, Elastix B-spline) are reported as unavailable due to convergence failure during inversion of the corresponding B-spline transform. Code Python scripts used to generate the figures reported in the manuscript: Bland–Altman and per-patient Dice comparison figures — computes agreement statistics (mean difference, standard deviation, and 95% limits of agreement) between ANTs SyN and Elastix B-spline under cavity-masked cost-function masking, and produces (1) Bland–Altman plots for Dice and HD95 showing per-patient agreement with mean bias and 95% limits of agreement, and (2) a paired per-patient comparison plot of Dice scores for both methods. Summary statistics are printed alongside each figure so results can be verified independently of the plots. [Figure 3/5/6/7 script — one line here on what it computes/plots, so a reader knows what to expect without opening the code]