Skip to Content

AI and Synthetic Data For Gait Analysis in Neurology

Unlocking the Secrets of Our Stride

Every step we take carries hidden information about our health. Gait, the way we walk, provides subtle clues that can indicate early stages of neurological disorders such as Parkinson’s disease, cerebral palsy, and dementia. 

While traditional assessments rely heavily on visual observation, these methods can be subjective and inconsistent. Today, artificial intelligence is poised to transform gait analysis, making it more objective, scalable, and accessible both within clinics and beyond.

The Power of AI and Synthetic Data

A pioneering collaboration between IBM Research and Cleveland Clinic has produced GaitFM, an innovative AI model trained to understand and analyze walking patterns. What sets GaitFM apart is its foundation: thousands of synthetic gait sequences crafted through advanced physics-based simulations. 

These simulations replicate a diverse range of walking styles, influenced by variables like age, medical conditions, and the type of sensors used. Unlike earlier models constrained by limited datasets, GaitFM is engineered to adapt to different populations and technologies including commonplace devices like smartphones and wearables.

Why Synthetic Data Transforms the Field

Real clinical gait data can be difficult and slow to collect, particularly for rare conditions or underrepresented groups. Synthetic data addresses this challenge by offering:

  • Diversity: A broad spectrum of gait variations that account for different diseases, body types, and sensor setups.

  • Realism: Biomechanically accurate simulations ensure clinical relevance.

  • Scalability: The ability to rapidly incorporate new diseases, edge cases, or emerging devices.

This strategy helps AI models overcome practical hurdles, such as inconsistent camera angles or the variability of wearable devices, which have limited prior approaches.

Blending Synthetic and Real-World Data

GaitFM’s training begins with synthetic data and is further refined using real patient data from those with neurological conditions. This hybrid approach delivers high accuracy in estimating gait parameters and predicting clinical outcomes, even with far less real-world data than previously required. Key strengths of the model include:

  • Data-Efficient Learning: Synthetic pre-training reduces the need for vast clinical datasets while maintaining strong performance.

  • Multimodal Flexibility: GaitFM operates with various sensors (video, accelerometers, and EMG) enabling use in both professional settings and at home. Remarkably, it can estimate muscle activity from a single video feed, eliminating the need for specialized equipment.

This adaptability is vital for extending advanced gait analysis to broader communities and conditions that have historically lacked sufficient data.

Real-World Applications and Future Prospects

GaitFM’s excellence has already been recognized with the Best Paper Award at the 2025 IEEE International Conference on Digital Health. Ongoing pilots at Cleveland Clinic are testing the system in real clinical environments, with plans to gather even more data to refine its accuracy and resilience. 

The ultimate aim is to make neurological monitoring more equitable and accessible, using everyday technology like smartphones and wearables alongside synthetic data to keep the system robust and up to date.

By transforming ordinary movement into actionable clinical insights, this technology could revolutionize how neurological conditions are detected and managed, opening doors to earlier diagnosis and better treatment monitoring for more people.

The Bottom Line: Smarter, More Inclusive Gait-Based Care

IBM and Cleveland Clinic’s work signals a new era in digital health. Leveraging AI and synthetic data, clinicians can now gain deeper, more reliable insights into walking patterns, fostering earlier intervention and more personalized care for neurological disorders. The future of gait analysis is here and it’s more dynamic, adaptable, and inclusive than ever before.

Source: IBM Research Blog | Paper

AI and Synthetic Data For Gait Analysis in Neurology
Joshua Berkowitz October 3, 2025
Views 1045
Share this post