Cadaveric Ankle Ligament Strains During Robot-Driven Stair-Climbing With In Vivo Data-Trained AI Models

PURPOSE: This study measured dynamic in vitro strains in the lateral ankle ligaments during stair-climbing exercises (SCE) driven by a robotic system with AI-predicted joint kinematics. METHODS: Three-dimensional (3D) motions of the tibia, talus, and calcaneus during SCE were measured from fourteen healthy adults using a validated method registering 3D CT-based bone models to 2D biplane fluoroscopy images. The ankle morphology was described by parameters from a statistical shape model (SSM) trained by 74 cadavers. These in vivo kinematic and morphological data were used to train and validate a fully connected neural network (FCNN) to predict 3D joint kinematics during SCE from morphological data. The accuracy of the model's predictions was quantified by root-mean-squared errors (RMSEs). Three cadaveric ankles were CT scanned and input into the validated FCNN to get their kinematic data during SCE, which were used to drive each specimen to perform SCE via a robotic system. The dynamic strains of the anterior talofibular ligament (ATFL), calcaneofibular ligament (CFL), and posterior talofibular ligament (PTFL) were measured using a digital image correlation method. RESULTS: The developed SSM had a high accuracy with a mean RMSE of 0.5 mm. The mean RMSEs of the FCNN predictions were 0.9 mm, 2.1 mm, and 2.0 mm for the anterior/posterior, proximal/distal, and medial/lateral translations, respectively, while those for the sagittal, frontal, and transverse rotations were 3.3°, 1.6°, and 0.8°, respectively. The dynamic ATFL strains showed a greater peak value than those of the CFL and PTFL (Fig. 1). CONCLUSION: This study proposed a new AI-based approach to predict joint kinematics during SCE to drive cadaveric ankle joints via a robotic system for measuring dynamic strains in the ligaments. The proposed approach achieved reasonably high accuracy in predicting personalized joint kinematics and showed great potential for measuring ligament strains during functional exercises.