Affecting 5-10% of infants worldwide, Cerebral Palsy (CP) is the most common motor disability among infants, presenting as a variety of motor, perception, and cognitive impairments.
CP is a lifelong condition with no cure, but treatment exists through rehabilitation. Evidence suggests that rehabilitation can be especially effective before the age of 2, but to deliver it during this essential window of development, early detection is imperative. There are clinical assessments for early detection, but they often require extensive and time-consuming clinical training, and can only be administered by specialists, not primary care physicians. These barriers have resulted in low implementation in low-resource settings, prompting a growing demand to develop technology-assisted tools for easier, faster, and more objective testing.

One such quantitative measure is the Center of Pressure (COP), a metric capable of quantifying postural control. As postural control reflects balance and motor skill, COP shows promise in its ability to distinguish between typically developing and at-risk infants. Although COP has made it easier to detect impairment, force plates are not always accessible in low-resource settings due to price and a general lack of portability. To overcome these limitations, there are efforts to predict COP using cameras, as they are highly accessible and portable. Previous studies have utilized deep learning algorithms to extrapolate COP from pose estimation, relying exclusively on data-driven modeling to fill in the lack of dynamic information. Despite these efforts, no study has sufficiently predicted COP for clinical use or for infants.

This study proposes a solution to this issue by developing the first physics-based infant COP prediction model. Although a physics-informed model would output a predicted COP, there might still be some generalization based error. To offset the error, this study also introduces a hybrid physics-data-driven model, integrating a deep learning model to refine the results of the developed computational model. Through the combined use of physics and data-informed modeling, this study aims not only to accurately predict COP but also to better understand the relationship between infant motion and postural stability.