2D/3D Registration of Dynamic MRI for In vivo Knee Kinematics under Physiological Loading

The knee is particularly prone to ligament injury and osteoarthritis, yet static imaging often fails to capture functional joint behavior under physiological loading. Misalignment between static and dynamic assessments can lead to diagnostic inaccuracies, highlighting the need for in vivo motion evaluation. This study presents a non‑invasive framework combining weight‑bearing 0.25 T MRI, an MR‑compatible flexion‑guiding device, and automated 2D/3D registration to quantify in vivo knee kinematics under physiological loading.

The proposed method was evaluated using two MRI datasets: a set of 5 static, multislice 3D scans, to provide a reference kinematics, and single-plane 2D real-time images, used to reconstruct in vivo knee kinematics during the whole flexion cycle. The continuous knee kinematics were obtained by registering each 2D real-time image to an interpolated slice generated from the 3D bone models. The registration employed a similarity metric combining Gradient Differences with a penalty term that guarantees physiological femur–tibia articular contact. Measurement errors, defined as the differences between the static (3D) and continuous (2D) kinematics, were used to quantify the registration performance.

The developed 2D/3D registration algorithm reconstructed in vivo knee kinematics from single-plane dynamic MRI (DMRI) with a mean rotation error of 1.83° ± 1.47° and mean translation error of 2.26 mm ± 0.66 mm. Robustness testing across scanning planes identified the axial plane as optimal.

The proposed method achieved accuracy consistent with previous DMRI‑based approaches. Compared with dual‑plane acquisition on the same 0.25 T scanner, it preserved comparable rotational accuracy while exhibiting reduced accuracy to out‑of‑plane translations, an expected trade-off of single‑plane imaging. Accuracy is inherently limited by low‑field real‑time MRI, whereas higher‑field systems report superior performance only under static, non-weight‑bearing conditions. Reconstructed kinematics captured key physiological features, including tibial internal rotation and femoral rollback. Despite residual errors, the single‑plane framework remains encouraging and is easily generalizable to other joints.