UCSF

Objectives: To develop an automated intra-patient atlas based bone segmentation technique that is reliable for the quantification of in-vivo MRI knee kinematics.

Materials and Methods: 30 patients who had unilateral anterior cruciate ligament (ACL) tear and under went ACL single-bundle reconstruction received loaded-MRI scans. The same procedure was done for 6 healthy control patients for reproducibility. Femur and tibia were segmented automatically, and using in-house MATLAB software, three-dimensional segmentations were used to obtain kinematic data (Internal Tibial Rotation [ITR] and Tibial Position [TP]). Automatic segmentation and kinematic data was compared to results from previously used semi-automatic segmentation methods.

Results: Automatic segmentation was successful in 98.9% of 175 cases tested, with high similarity between masks of automatic and semi-automatic segmentation (85.1%). Automatic segmentation algorithm showed excellent reproducibility results between scan and rescan segmentations with an average absolute difference of: TP: 0.46 +/- 0.41 mm, ITR: 1.60 +/- 1.58 deg. These values were superior to those obtained semi-automatically. Automatic and semi-automatic kinematic results showed high correlation for both TP and ITR, with R=0.935 and R=0.900 respectively. [add summary of longitudinal changes of TP and ITR in patients]

Conclusions: The automatic segmentation technique developed in this paper proves to be very useful in reducing segmentation time and labor as well as in reducing user-generated bias. The proposed technique demonstrates a highly consistent and robust method that is useful for longitudinal knee kinematics quantification for in-vivo MRI.