Background:
Accurate, low-burden methods for quantifying real-world gait are needed in neuromuscular conditions such as Duchenne muscular dystrophy (DMD). Three recent studies using single-sensor accelerometry, collected entirely from smartphones provide complementary advances relevant to scalable, real-world monitoring. Here we integrate the studies and outline how they support future STRIDE-AI implementation.
Methods and Findings:
The first study applied smartphone waist accelerometry to characterize gait in boys with DMD and typically developing peers. Machine-learning and deep-learning models distinguished DMD from typical gait using temporospatial gait features, demonstrating that smartphone devices can capture clinically relevant gait signatures.
The second study evaluated wearable-sensor calibration across slow, self-selected, and fast walking. Machine learning-based step detection combined with regression models accurately estimated stride length, walking velocity, and travel distance across a range of gait speeds, showing that data-driven calibration performs well even in populations with atypical biomechanics.
The third study developed a simplified frequency-domain estimator for travel distance. Using a single waist-worn accelerometer, Fast Fourier Transform (FFT) analysis identified dominant step-frequency peaks; when paired with subject height and regression modeling, step length and distance were estimated accurately across walking and running speeds. This demonstrated the feasibility of a low-power, FFT-centered pipeline for mobile deployment.
Integration and Future Directions:
These findings support a unified mobile mobility-analysis pipeline integrating: (1) ML-based step-event detection and stride-length calibration, (2) frequency-domain gait features for pathological gait classification, and (3) a lightweight FFT module for community step and distance estimation. STRIDE-AI will incorporate these components with geolocation and ecological momentary assessment Quality of Life tools to enable passive, everyday longitudinal real-world mobility behaviors monitoring using smartphones.
Conclusion:
Integrating ML-based gait characterization with FFT-derived step-frequency analysis offers a scalable, mobile-ready pathway for real-world gait assessment and will form a core component of future STRIDE-AI tools.