Aims: To develop and evaluate a novel AI-based video analysis tool for the quantitative assessment of “Places Pegs in” and “Blue Board” tasks in the Bayley Scales of Infant Development (BSID-II). Methods: A prospective cohort study was conducted from February 2022 to December 2022, including children aged 12–42 months referred for suspected developmental delay. Participants were evaluated using the BSID-II, and their performances on the two tasks were video recorded and analyzed with the novel tool. Sensitivity and specificity were determined by comparing the tool’s results to standard BSID-II assessments by therapists. Data collected included total time, number of trials, successful trials, and time and spatial intervals for each trial. Children were classified into typically developing (TD) (MDI ≥ 85) and developmental delay (DD) (MDI < 85) groups based on their mental developmental index (MDI). Results: A total of 75 children participated in the study, and the mean values of MDI and PDI for the enrolled children were 88.9 ± 18.7 and 80.0 ± 16.7. The “Places Pegs in” had 86.5% sensitivity and 100% specificity; the “Blue Board” had 96.9% sensitivity and 89.5% specificity. Differences in cumulative successes over time were observed between age groups and TD and DD groups. The tool automatically calculated maximum successes at specific time points. Interpretation: The AI-based tool showed high predictive accuracy for BSID-II tasks in children aged 12–42 months, indicating potential utility for developmental assessments.
Camera-based sensing may provide a scalable approach to quantifying hand function in children with cerebral palsy (CP), but fine grasp patterns remain challenging because of occlusion, small anatomical structures, and complex hand–obobject interactions. This study investigated a single-camera AI-based framework for classifying the developmentally ordered grasp patterns of the Quality of Upper Extremity Skills Test (QUEST). Grasp-of-Cereal and Grasp-of-Cube tasks performed by 41 children with CP were recorded with a single RGB camera and analyzed using MediaPipe hand landmarks, SAM2-based hand segmentation, YOLOv9 object localization, and a 3D DenseNet-121 classifier. Trial-level accuracy, severity-weighted accuracy (SWA), and quadratic weighted kappa (QWK) were estimated with participant-level bootstrap 95% confidence intervals (CIs). Accuracy was 0.84 (95% CI, 0.750–0.914) for Cereal and 0.85 (0.795–0.901) for Cube; SWA was 0.922 and 0.907, and QWK was 0.579 and 0.756, respectively. Therapist inter-rater QWK (0.883 and 0.838) is reported as a descriptive reference only. Cereal macro-F1 (0.534) and balanced accuracy (0.494) indicated uneven class performance. Development and test recordings were temporally separated but not participant-independent; for Cube, 39 of the 41 children contributing development data also contributed follow-up test recordings. These findings support the feasibility of longitudinal within-cohort grasp classification but do not establish generalization to previously unseen children.