# Machine learning-based estimation of structure-specific load around the ankle and knee joint during running using IMU data

## TL;DR

This paper presents a machine learning model that estimates musculoskeletal loads on the ankle and knee during running using IMU data, offering a faster and more practical alternative to traditional methods.

## Contribution

A novel LSTM-based model is introduced to estimate structure-specific loads using IMU data, achieving high accuracy with minimal sensors.

## Key findings

- The model achieved R² scores of 0.84–0.93 for estimating loads on tendons and joints during running.
- SSL can be reliably estimated using only a single IMU mounted on the pelvis.
- The model shows potential for real-world applications in injury prevention and rehabilitation.

## Abstract

Running imposes substantial repetitive loads on the musculoskeletal system, which can lead to running-related overuse injuries. Therefore, effective structure-specific load management is essential for both prevention and rehabilitation. However, the conventional method for estimating structure-specific load (SSL) is time intensive to execute, and the resulting data is computationally expensive to analyze. This study aims to estimate the SSL on the Achilles and the patellar tendons, and the knee and ankle joint during running from data collected by one or two inertial measurement units (IMUs). We proposed a long-short-term-memory-based model that is trained and evaluated on a dataset of 43 participants. The estimated SSL during the stance phase of a running step achieved
R2
scores of 0.93, 0.84, 0.89, and 0.86 compared to the ground truth values for the Achilles and patellar tendon force, and ankle and knee contact forces, respectively. Furthermore, the results indicate that the SSL can be estimated practically and reliably during running using only time series data from a single pelvis-mounted IMU. These findings highlight the potential of using machine learning models applied to IMU data for SSL monitoring in real-world running scenarios.