Mobile Stride Length Estimation With Deep Convolutional Neural Networks.

Area of Science:

• Biomedical Engineering
• Neurology
• Rehabilitation Science

Background:

• Accurate spatial gait characteristic estimation is vital for assessing motor impairments.
• Current double integration methods for gait analysis are limited by methodological constraints, particularly the need for a clear zero-velocity phase.

Purpose of the Study:

• To develop and evaluate a novel approach for stride length estimation using deep convolutional neural networks.
• To overcome the limitations of existing methods in clinical gait analysis.

Main Methods:

• A deep convolutional neural network was developed to map stride-specific inertial sensor data to stride length.
• The model was trained on a benchmark dataset of 1220 strides from 101 geriatric patients.
• Evaluation involved tenfold cross-validation using three different stride definitions.

Main Results:

• The novel method achieved high accuracy and precision in stride length estimation, outperforming state-of-the-art methods.
• Performance was robust across different stride definitions, with best results achieved using midstance to midstance definitions.

Conclusions:

• The proposed deep learning approach offers a more versatile and precise method for mobile stride length estimation.
• This method is not constrained by the need for a zero-velocity phase, expanding its clinical applicability for gait analysis in various neurological and musculoskeletal conditions.