A Dataset of Clinical Gait Signals with Wearable Sensors from Healthy, Neurological, and Orthopedic Cohorts

Open access, clean, annotated databases are key for future significant advances in gait quantification with inertial sensors. This multi-pathology and clinically annotated dataset provides 1356 gait trials from 260 participants equipped with four inertial measurement units placed on the head, lower back, and dorsal part of each foot. Participants followed a standardized protocol: standing still, walking 10 meters, turning around, walking back 10 meters, and stopping. It results in a large human walking dataset with over 11 hours of gait time series data. The quality is ensured by the documentation and metadata provided. The study population encompasses healthy individuals and patients with neurological (parkinson disease, cerebrovascular accident, radiation-induced leukoencephalopathy and chemotherapy-induced peripheral neuropathy) or orthopedic (hip osteoarthritis, knee osteoarthritis and anterior cruciate ligament injury) conditions. For each pathology, the most relevant clinical or radioclinical score has been calculated to provide insight into the gravity of the disease. This dataset can be used to study kinematic parameters, gait cycles time series, and various indicators for quantifying gait in routine clinical practice.

The study of gait analysis has increased exponentially in medical and preventive interest due to its critical role in understanding various physiological functions and pathologies 1 . Gait involves complex systems, making it particularly susceptible to alterations in a wide range of medical specialities, including neurological 2,3 , orthopedic 4 , cardiac 5 , pulmonary 6 , and even oncological 7 conditions. Consequently, the quality and reliability of instrumental gait analysis is a significant challenge and opportunity for the progression of medical practice.

Traditionally, quantitative gait analysis has been conducted in dedicated laboratories using sophisticated motion capture systems. However, the development of inertial measurement units (IMUs) has revolutionized this field by offering a more accessible, lighter, and cost-effective alternative 8,9 . Numerous validation studies have shown IMUs can provide equivalent accuracy in detecting gait kinematics traditional motion capture systems 10,11 . As a result, IMUs have democratized gait analysis, enabling a broader application in both clinical and research settings.

The transition to using IMUs exclusively for gait analysis has highlighted the importance of sharing high-quality datasets with precise clinical metadata. These datasets often include annotations with event detection and gait characteristics recorded by gold-standard methods (e.g., motion capture, force platforms, pressure-sensitive treadmills). Such annotated datasets are crucial for training algorithms to detect events and calculate gait parameters using IMUs, thus validating their effectiveness 12-17 . Furthermore, datasets without these annotations are valuable for the evaluation and learning of gait parameters, provided they contain