AI-Based Video Analysis for Motor Development Assessment in Children

Artificial Intelligence-Based Movement Analysis, Motor Development Assessment

Motor Development, Video Analysis, Artificial Intelligence, Deep Learning

This is a non-interventional, prospective observational study aimed at developing and validating an artificial intelligence-based system for assessing motor development in children using video analysis. Children aged 5 to 10 years will perform standardized motor tasks, which will be recorded under controlled conditions. The recorded videos will be analyzed using computer vision and deep learning techniques to extract movement patterns. The results of the AI-based analysis will be compared with standardized motor assessment scores obtained from the Bruininks-Oseretsky Test of Motor Proficiency, Second Edition - Short Form (BOT-2 SF). Participants will be classified into typical and atypical motor development groups based on BOT-2 scores. The primary objective is to evaluate the classification performance of the AI model. Secondary analyses will examine the relationship between AI predictions and continuous motor performance scores. The study is designed to explore whether motor development can be assessed objectively without direct clinical testing, using only short video recordings. The findings may contribute to the development of scalable and accessible digital screening tools for early identification of motor development differences in children.

This study is a prospective, non-interventional observational study conducted to develop and validate an artificial intelligence-based system for the assessment of motor development in children. The study includes children aged between 5 and 10 years who have no previously diagnosed neurological, developmental, or orthopedic disorders. All participants will complete the Bruininks-Oseretsky Test of Motor Proficiency, Second Edition - Short Form (BOT-2 SF), which will serve as the reference standard for motor performance. Based on BOT-2 scores, participants will be categorized into typical and atypical motor development groups using predefined thresholds derived from normative data and statistical distribution methods. In addition to standardized testing, participants will perform a series of structured motor tasks, including jumping jacks, tandem walking, skipping, single-leg balance, finger-to-nose coordination, and protective extension responses. These tasks will be recorded using high-resolution video under controlled environmental conditions. Video data will be processed using computer vision pipelines. Skeletal keypoints will be extracted using pose estimation models, and silhouette segmentation will be obtained using deep learning-based segmentation models. Extracted features will be normalized and used as input for machine learning and deep learning architectures, including transformer-based models and graph-based networks. The primary outcome is the classification performance of the AI model in distinguishing typical versus atypical motor development profiles, evaluated using metrics such as ROC-AUC, accuracy, sensitivity, specificity, F1-score, and balanced accuracy. Secondary outcomes include regression performance for predicting continuous motor scores, evaluated using MAE, RMSE, and R-squared values. Inter-rater reliability of expert evaluations will be assessed using intraclass correlation coefficients (ICC). Additional analyses will include error distribution examination and Bland-Altman analysis to assess agreement between AI predictions and standardized test scores. This study does not involve any intervention, treatment, or risk beyond standard observational procedures. All participants are healthy volunteers, and informed consent will be obtained from parents or legal guardians. The study has been approved by the Istanbul Medipol University Non-Interventional Clinical Research Ethics Committee.

No related papers found yet.

Turkey (Türkiye)