Intelligent wearable allows out-of-the-lab tracking of developing motor abilities in infants

Early neurodevelopmental care lacks objective and scalable assessment tools for infants' motor abilities. Current methods, often subjective and performed in controlled settings, hinder generalizability. This study aimed to develop and validate a method using wearable technology to measure spontaneous motor abilities in infants across all developmental milestones, from lying supine to walking. The importance of this research stems from the global challenge of providing effective early neurological assessments. Early detection of motor delays is crucial for timely interventions, improving long-term outcomes. While developmental milestones are useful for population screening, they lack the sensitivity to capture the nuanced variability of natural motor development. Existing standard assessments, though providing fine-grained information, are often subjective and conducted in artificial environments. This study addresses the need for objective, ecologically valid methods for tracking neurodevelopment by leveraging advances in sensor technology to record and quantify infants' spontaneous motor activity in real-world settings.

The literature highlights the challenges in objectively assessing infant motor development. Studies demonstrate the need for scalable and reliable tools to track motor milestones and spontaneous movements. Existing clinical assessments, while providing valuable insights, often rely on subjective observations and are limited by the controlled environment of the clinic. Several studies have explored the use of sensor technology to capture infant movements, but these are often limited by data availability, recording conditions, and the number of sensors used. The lack of a comprehensive, easily interpretable system for classifying infant movements has also hindered progress. This research builds upon prior work by proposing a new framework that combines advanced wearable technology with deep learning techniques to overcome these limitations.

This study employed a cross-sectional design, recruiting 59 infants (5-19 months) from Helsinki University Hospital. Participants included healthy infants, those with mild perinatal asphyxia, and preterm infants. A novel wearable device, MAIJU, was developed. MAIJU is a jumpsuit equipped with four multi-sensor units (Movesense) recording tri-axial linear acceleration and angular velocity at 52 Hz. Data were wirelessly transmitted to a mobile application. Recordings lasted 18-199 minutes (average 67 minutes), conducted either at home or in a research lab designed to mimic a home environment. A new motor ability description scheme was created, classifying postures (prone, crawling, sitting, standing, side-lying) and movements (still, proto, elementary, fluent) for each second of recording. Two annotators independently coded video recordings from a subset of participants. A deep learning-based convolutional neural network (CNN) was trained using the annotated data to classify postures and movements, achieving human-equivalent accuracy. Finally, a novel neurodevelopmental index, BIMS, was trained using the aggregated data from the wearable to estimate the maturity of infant motor abilities and was compared to the Alberta Infant Motor Scale (AIMS) and parental surveys.

The MAIJU wearable successfully recorded data from all participants. The deep learning-based classifier showed human-equivalent accuracy in classifying postures and movements, with high inter-rater reliability (kappa > 0.93 for postures and 0.60 for movements). The BIMS score, derived from the aggregated MAIJU data, strongly correlated with chronological age (Pearson's r = 0.89, p < 1e-20), with a mean absolute error of 1.6 months. BIMS also showed a strong correlation with the AIMS score (r = 0.83, Spearman's rho = 0.82), demonstrating its clinical relevance. Comparison with parental surveys revealed discrepancies, highlighting the benefits of objective measurement. The analysis using self-supervised learning confirmed that the identified motor ability categories were genuinely present in the MAIJU data.

This study demonstrates the feasibility of using a wearable sensor system for accurate and objective assessment of infant motor development in real-world settings. The high accuracy and strong correlation with established clinical measures suggest that the BIMS score is a valuable tool for neurodevelopmental assessment. The method overcomes limitations of previous approaches, providing a scalable and cost-effective method suitable for both clinical practice and research. The objective nature of the assessment reduces reliance on subjective interpretation and inter-rater variability. The ability to perform assessments outside of the clinic enhances ecological validity and accessibility. The combination of wearable technology and machine learning offers a powerful approach to early identification and intervention for neurodevelopmental delays.

This study successfully developed and validated a wearable system and analytical approach for objective and scalable assessment of infant motor abilities. The BIMS score provides a novel, reliable metric for tracking motor development, with potential applications in early diagnosis, intervention, and research. Future research should focus on longitudinal studies to further evaluate the clinical utility and predictive validity of the BIMS score across diverse populations and clinical contexts. Larger, more diverse datasets are needed to refine the motor ability classification scheme and improve the generalizability of the BIMS score.

The study's cross-sectional design limits the ability to draw conclusions about the longitudinal trajectory of motor development. The sample size, while substantial, may not fully represent the diversity of infant populations. The study predominantly used data collected in Finland, potentially impacting the generalizability of the findings to other cultural contexts. Further research is needed to evaluate the performance of the system across a broader range of developmental conditions and diverse cultural settings.