Wearable and Deep Learning-Based Recognition of Real-World Movement Behavior of Adolescents With Cerebral Palsy: Feasibility and Discriminant Validity Study

Cerebral palsy (CP) is a non-progressive disorder resulting from injuries or abnormalities in fetal or early infant brain development. According to registries from European countries, the condition affects 2-3 out of every 1000 live births. An individual with CP typically presents with motor development disorders that cause abnormal patterns of movement and posture due to impaired coordination of movements and muscle tone regulation. People with cerebral palsy can also have various other problems, including sensory and cognitive problems and sleep disturbances. These symptoms result in limitations in activity level and societal participation throughout the individual's life. Adolescents and even children as young as seven may experience a decline in motor ability, leading to changes in their movement behavior. Healthcare professionals rely on various observations and measurements performed in clinical and hospital settings to assess and treat individuals with CP. However, there is some uncertainty about whether these assessments truly reflect real-life movement behaviors, as using an impaired extremity in everyday life frequently deviates from its motor capacity. There is an absence of robust tools that capture daytime and nighttime movement behavior in real-world settings rather than in clinical or controlled environments. Hemiparesis is the most common marker of CP, making asymmetrical deficits a target for intensive interventions such as physical and occupational therapy. Yet, no clinical tools are available that document asymmetrical differences in the real world in children and adolescents with CP. An objective method to measure real-world movement patterns would allow therapists to identify individuals who need a more comprehensive evaluation and to target interventions and other management strategies more precisely. This would help children and adolescents with CP gain motor skills to maximize independence. Further, objectively observing individuals with CP in their daily lives is essential to gain insights into functional decline. It has been observed that children and adolescents with CP are more likely to experience sleep-related difficulties such as difficulty initiating sleep, frequent nocturnal awakenings, discomfort while in bed, and early morning awakenings. As sleep quality plays a vital role in health-related quality of life, it is crucial to have objective methods to evaluate and monitor potential sleep problems in a real-world context.

A deep-learning convolutional neural network has been modeled to recognize postures lying, sitting, and standing the activity of walking, and movements of the right and left extremities. The network uses accelerometer and gyroscope data from 7 wearable sensors. Testing of the network´s performance found that it surpasses human annotators in accurately classifying the movement behavior of healthy and typically developed adults. These findings are currently under review and have yet to be published. The present protocol details the methodology for assessing the feasibility of real-world movement behavior monitoring and the discriminant validity of the network in adolescents with CP and controls.

The feasibility evaluation examines the technology used, e.g., potential data loss and the credibility of data output, as well as user acceptance, e.g., sensor wear time and adverse events. The networks' discriminant ability will be assessed by the network's ability to differentiate between controls and CP severity, e.g., scores on the Gross Motor Functional Classification Scale - Expanded and revised (GMFCS-E&R), different types of CP, differently affected body parts of the participating adolescents with CP, as well as individuals who have and have not sleep problems in the entire cohort.