An End-to-End Pipeline for Quantitative MRI Analysis of Ligament/graft Healing

Links

URL: https://repository.library.brown.edu/studio/item/bdr:jt9jt5y4/

Abstract

Anterior cruciate ligament (ACL) disruption is a common injury, particularly in the young and active patient. ACL reconstruction surgery is the current standard of care, yet reinjury rates are high, and it does not significantly reduce the risk of post‐traumatic osteoarthritis (PTOA) relative to non‐operative management. A prerequisite for improving interventions is the development of robust measures of ACL healing that directly measure structural properties of the healing ligament or graft. A potential solution is quantitative magnetic resonance imaging (qMRI), which measures soft tissue structural properties, tissue degeneration, and relates to long‐term risk factors.

Objective measures of ACL healing will enable the identification of modifiable risk factors to improve surgical outcomes. Furthermore, it will allow for the comparison of intervention methods with a shorter feedback loop than long‐term outcomes tracking (e.g., registry data). However, widespread adoption of qMRI has been limited outside of research settings, due to challenges such as the inter‐ scanner reproducibility of qMRI sequences, and the time and expertise requirements associated with post‐processing steps such as image segmentation.

The primary goal of this work was to create and validate an automated qMRI pipeline that addresses the key technical bottlenecks for qMRI of the ACL. First, the inter‐scanner reproducibility of two common qMRI sequences (T₂* relaxometry and CISS) was assessed, and post‐acquisition harmonization methods were evaluated to address the observed differences. Second, a deep learning‐ based image segmentation method for the ACL was developed, which automates the most time‐ consuming post‐processing step for qMRI and eliminates the inter‐ and intra‐segmenter variability inherent to manual image segmentation. Third, it was shown that common qMRI features relate prospectively to functional outcomes post‐ACL surgery. Fourth, an improved T₂* relaxometry‐based machine learning model for ACL failure load prediction was developed. Finally, the processing components validated in Aims 1-4 were integrated into a single platform to enable fully automated post-processing of qMRI data. By increasing the practicality of qMRI, the platform created in this project aids in the development of more effective interventions for ACL injury.

Cited Works (5)

Year	Entry
2008	Renstrom P, Ljungqvist A, Arendt E, Beynnon B, Fukubayashi T, Garrett W, Georgoulis T, Hewett TE, Johnson R, Krosshaug T, Mandelbaum B, Micheli L, Myklebust G, Roos E, Roos H, Schamasch P, Shultz S, Werner S, Wojtys E, Engebretsen L. Non-contact ACL injuries in female athletes: an International Olympic Committee current concepts statement. Br J Sports Med. June 2008;42(6):394-412.
2022	Flannery SW, Walsh EG, Sanborn RM, Chrostek CA, Costa MQ, Kaushal SG, Murray MM, Fleming BC, Kiapour AM. Reproducibility and postacquisition correction methods for quantitative magnetic resonance imaging of the anterior cruciate ligament (ACL). J Orthop Res. December 2022;40(12):2908-2913.
2000	Boden BP, Dean GS, Feagin JA Jr, Garrett WE Jr. Mechanisms of anterior cruciate ligament injury. Orthopedics. June 2000;23(6):573-578.
2011	Anderson DD, Chubinskaya S, Guilak F, Martin JA, Oegema TR, Olson SA, Buckwalter JA. Post-traumatic osteoarthritis: improved understanding and opportunities for early intervention. J Orthop Res. June 2011;29(6):802-809.
2018	Samelson EJ, Demissie S, Cupples LA, Zhang X, Xu H, Liu C-T, Boyd SK, McLean RR, Broe KE, Kiel DP, Bouxsein ML. Diabetes and deficits in cortical bone density, microarchitecture, and bone size: Framingham HR-pQCT study. J Bone Miner Res. January 2018;33(1):54-62.