[home] [Personal Program] [Help]

---

Session: Poster session 2 (Odd numbers)
Session starts: Friday 27 January, 10:00
Presentation starts: 10:00

---

Sanchana Krishnakumar (University of Twente)
Bert-Jan van Beijnum (University of Twente)
Jaap Buurke (University of Twente)
Peter Veltink (University of Twente)
Chris Baten (Roessingh Research And Development)

Abstract:
Rehabilitation after an anterior cruciate ligament (ACL) injury is divided into multiple phases and progress between phases is based on the functional assessment of patients. These assessments are currently subjective and are done by visual monitoring of activities such as running, hopping, jump landing, and cutting maneuvers by physiotherapists. Estimation of objective measures like knee joint moments and ground reaction forces (GRF) during assessment can help in gaining new insights on knee loading and open new avenues for rehabilitation. Accurate estimation of kinetics is a complex task and requires expensive motion capture systems along with force platforms. On the other hand, several algorithms have been proposed in the literature to estimate kinetics by just using kinematics measured with inertial sensors (IMUs). However, the knowledge about their clinical applicability is limited. This study aims to compare available algorithms for the prediction of GRF and/or estimation of joint moments only using IMUs and evaluate their feasibility for adoption in ACL rehabilitation. A literature search was conducted using Scopus (Elsevier), PubMed and SportDiscus with keywords including ‘GRF’, ‘Joint kinetics’, ‘IMU’ and commonly accepted variations of these terms. The identified studies were classified based on the parameters estimated (joint moments/GRF) and the principles used such as machine learning (ML), musculoskeletal modelling, hybrid, direct modelling, or statistical approach. The comparison of the algorithms was done based on the accuracy achieved, assumptions used, tasks validated and their applicability for ACL patients. Most of the studies evaluated have estimated only vertical GRF with good accuracy and reported lateral GRFs as less reliable. ML-based approaches have proved to be more versatile but have the disadvantage of sensitivity to input parameters and require large sets of training data. Tasks such as walking, and landing also involve double support phases where further transfer functions are required to distribute forces between the legs. The applicability of assumptions made for distribution is unclear for ACL patients. The assessed algorithms have also not yet been widely validated for tasks such as jump landing and hoping. A combination of two methods such as biomechanical modelling and ML or musculoskeletal modelling may be used to further increase accuracy and make them versatile to estimate joint moments for a large range of movements. Further validation and tuning of these algorithms are thus necessary before being implemented for ACL monitoring and phase decision.