Integrating deep learning in stride-to-stride muscle activity estimation of young and old adults with wearable inertial measurement units.

Deep learning has become powerful and yet versatile tool that allows for the extraction of complex patterns from rich datasets. One field that can benefits from this advancement is human gait analysis. Conventional gait analysis requires a specialized laboratory setup with markers positioned at anatomical landmarks. It also demands accurate muscle identification and electrode placement. This work overcomes these limitations by proposing a Convolutional Neural Network (CNN) that processes tokenized data measured by the wearable Inertial Measurement Units (IMUs) to estimate muscle activities during walking. Gait data were gathered from 65 participants aged 19-73 years old. The proposed method was compared against the other deep learning models. The result of the proposed method was better that the alternatives. The results demonstrated that test dataset achieved Root Mean Squared Error (RMSE) less than 12% and Correlation Coefficient (r) greater than 90% for all muscles. It performed particularly well on unseen data, with most muscle activities showing RMSE below 20% and r above 80%. These findings indicate that the proposed technique can efficiently extract relevant gait data and estimate the muscle activity, potentially simplifying gait analysis and making it more affordable and accessible to a wider range of users.