The development and validation of motion tracking algorithms require realistic four-dimensional (4D) image datasets with accompanying ground truth (GT) motion information. However, such datasets spanning multiple respiratory cycles are scarce. This work presents a reproducible pipeline for synthesizing anatomically semi-realistic 4D motion phantoms with GT motion information, leveraging heterogeneous and publicly available data sources. Methods: The proposed approach synthesizes semi-realistic 4D image phantoms by integrating 4D deformation vector fields (DVFs) derived from publicly available 4D imaging datasets spanning a single respiratory cycle, static 3D target images, and multi-cycle respiratory traces. Continuous motion fields are produced by harmonizing these heterogeneous inputs through DVF alignment, spatial gap completion, and respiratory trace-guided DVF interpolation. These motion fields are then applied to deform the static target images, yielding temporally coherent, time-resolved 4D image sequences that capture respiratory motion over multiple cycles. Results: The proposed pipeline was evaluated by generating three distinct phantom datasets: Sets (a) comprise twenty 4D CT phantoms with multi-organ GT segmentations; Sets (b) consist of twenty 4D MR phantoms with multi-organ GT segmentations; Sets (c) include twenty 4D MR phantoms with GT DVFs. Quantitative evaluations demonstrated that local volumetric changes were physically plausible and that synthesized liver motion trajectories exhibited significant correlation (p < 0.01) with measured liver motion traces. Qualitative assessment by twelve experts yielded an average realism score of 4.2 on a five-point scale. Motion tracking experiments on the phantoms yielded performance comparable to real clinical sequences, suggesting that they provide a clinically representative evaluation setting. The generated image phantoms, source code, and input data were made publicly accessible to enable reproducibility and customization. Conclusion: The proposed pipeline and openly shared resource address a critical gap in medical imaging research by providing 4D image phantoms that capture physiological motion spanning multiple respiratory cycles. This work paves the way for more robust development, validation, and benchmarking of image-based motion tracking algorithms.
Elodie Lugez (Thu,) studied this question.