Comprehensive Summary
This study, conducted by Snyder et al., tested a neural network model predicting knee loads during various activities to help progress osteoarthritis research. The study focused on variables linked to knee osteoarthritis progression such as knee adduction moment (KAM), knee flexion moment (KFM) and medical joint contact force (MJCF) that usually require extensive resources to measure. The researchers built custom insoles with sensors and used motion capture to collect data from 47 participants performing different activities, including sit-to-stand, stand-to-sit, walking and running trials. The data collected was used to train recurrent neural network (RNN) models for each activity and variable, resulting in 12 models. Results indicate that MJCF and KFM predictions demonstrated great accuracy, with correlation coefficients ranging from 0.88 to 0.98 across all activities. However, KAM predictions showed partial accuracy with sit-to-stand and stand-to-sit activities, with coefficients of 0.50 ± 0.44 and 0.48 ± 0.40. The researchers state that the results are consistent with previous research using video based methods.
Outcomes and Implications
Knee osteoarthritis is an extremely common disease affecting people’s daily activities. However, the data necessary for research is difficult to obtain without professional equipment and trained personnel. The use of wearable technology such as insoles and neural network models could facilitate the data collection process. Previous studies have used neural networks to predict risk variables, however, they used video based methods which is difficult to be practical in daily settings. Snyder et al. emphasizes the need of future studies focusing on the impacts of reducing the risk factors of knee osteoarthritis. They believe that their research could provide a framework for future research to collect risk factor data outside a laboratory setting.