Strike index estimation using a convolutional neural network with a single, shoe-mounted inertial sensor.

Area of Science:

• Biomechanics
• Sports Science
• Machine Learning

Background:

• The strike index is a key metric for classifying running foot strike patterns.
• Traditional methods for strike index assessment require specialized laboratory equipment like force plates and motion capture systems.
• There is a need for accessible and accurate methods to measure strike index outside of clinical settings.

Purpose of the Study:

• To develop and validate a novel method for estimating the strike index using data from a shoe-mounted inertial measurement unit (IMU).
• To utilize a participant-independent convolutional neural network (CNN) for accurate strike index prediction.
• To enable reliable strike index assessment in non-laboratory environments.

Main Methods:

• Data collection involved 16 participants running with three distinct foot strike patterns (rearfoot, midfoot, forefoot) under varied conditions (footwear, speed).
• A convolutional neural network (CNN) architecture, including convolutional, max-pooling, and fully-connected layers, was employed for data analysis.
• The CNN model was trained and tested using data from the shoe-mounted IMU to estimate the strike index.

Main Results:

• The proposed method accurately estimated the strike index with a root mean square error of 6.9% and an R-squared value of 0.89.
• The trained CNN model demonstrated robustness to variations in running speed.
• The participant-independent model showed reliable performance across different running conditions.

Conclusions:

• The developed approach enables accurate strike index estimation using wearable IMU sensors and CNN analysis.
• This method offers a viable alternative to traditional laboratory-based assessments for running gait analysis.
• The findings suggest potential for improving running performance and reducing injury risk through accessible gait monitoring.