Analyzing Gait Pattern as a Diagnostic, Prognostic, and Preventive Tool in Disease

Categories: “Diagnostics”, “Neurological Disorders“, “Computer Science

Reference #: 2019-049

OTC Contact: Zeinab Abouissa,  Phone: 202-687-2702, Email:


According to the National Council on Aging, falls (related to gait disturbances) are the leading cause of both fatal and non-fatal injuries and hospital admissions in older adults.  25% of Americans over age 65 fall every year and the total cost of fall injuries in the U.S. is around $67.7 billion yearly, 75% of which is covered by Medicare and Medicaid, all figures which are expected to soar as the number of Americans age 65 and older grows..  This also does not take into account the financial burden of falls among younger groups and sports related injuries, all of which account for billions of dollars and cause tremendous physical and emotional difficulties. 

Researchers in Georgetown University’s Department of Neurology in collaboration with Department of Computer Science developed a precise system and method for detecting and assessing diseases and risk of injuries by analyzing specific gait measurements using acceleration and sound, thus providing a noninvasive and accurate diagnostic and prognostic system for fall prevention.  This is achieved using artificial intelligence (AI) by establishing a movement pattern for normal vs. disease states that affect gait such as Parkinson’s disease, neuromuscular disorders, post-stroke spasticity, etc.   

The underlying technology is based a neural network with a recurrent neural network layer and a fully connected layer that receives sensor data indicative of an individual’s gait and outputs an assessment regarding the individual’s health.  The neural network is trained using training data indicative of abnormal gaits and normal gaits.  To analyze the training data and the sensor data, the recurrent neural network layer parses each piece of data into a series of windows and analyzes each window in series to generate a context vector characterizing each window and the previously analyzed windows.  The fully connected layer, having been trained to differentiate between normal gaits and abnormal gaits based on context vectors characterizing the training data, is used to generate a final assessment characterizing the user gate as normal or abnormal using one or more of the context vectors characterizing the sensor data. 

Current commercially available gait analyzers are generally complicated and bulky and need a well-equipped room in a clinical setting to analyze a patient’s gait. This had been a major challenge for vendors, as this limits the adoption of gait systems.   Other prior methods reported which determine a person’s gait characteristics (e.g., strike distance, foot lift, foot pressure, etc.) do so over a number of steps and identify abnormal gait characteristics by analyzing the average gait characteristics from the entire data set collected from the patient.  The problem with these methods is that degeneration of a person’s gait is a gradual progression that results in only a portion of the gait suffering from abnormalities.  For example, patients with certain muscle disorders often experience waddling gait where they lack sufficient strength in the pelvic girdle muscles and may turn their pelvis to drag the their foot.  Patients with cerebellar disorders (e.g., stroke, tumor, and degenerative disorders) typically develop wide-based gait to keep balance.  Parkinson’s disease patients develop rapid, small, shuffling steps and a tendency to run (festination).  Even clinically depressed patients have been found to have demonstrated a smaller push-off force in the posterior and downward directions.  Thus, analyzing an entire data set of a patient’s gait characteristics may not detect specific gait degeneration.  

The present technology can be specifically used as a diagnostic tool by physicians and for assessment of treatment response.  Furthermore this device can be used for real time monitoring of gait in athletes to prevent sports-related injuries early on.   With increasing the robustness of the neural network, this technology with its analysis method will have increasing accuracy as well as potential for further diagnostic and prognostic purposes and management of mood disorders (e.g. depression) as well as monitoring a patient’s response to treatment.   


Gholam Motamedi, M.D.
Ophir Frieder, Ph.D.
Cristopher Flagg, Ph.D.


U.S. Patent Application No. 16/889,642
PCT Patent Application No. PCT/IB20/055183