Authors: (including presenting author): :
Lam SW, Kwong YY
Affiliation: :
Prosthetic and Orthotic Department, Kowloon Hospital
Keyword 1: :
Artificial Intelligence
Keyword 2: :
Positional Plagiocephaly
Keyword 3: :
Helmet Therapy
Keyword 4: :
Cranial Remoulding Helmet
Introduction: :
Positional plagiocephaly depicted a congenital craniofacial deformity on growing skull by extrinsic force among infants. Orthotic helmet was suggested as one of the preservative interventions for promoting symmetrical head growth and preventing further occipitoparietal flattening. Yet, inadequate research on intervention duration were investigated. The long period use of helmet, however, was reluctant to patients owing to discomfort or cosmetic concern. Providing personalized orthotic intervention duration to plagiocephaly patients, artificial intelligence (AI)-based model was computed on the prediction of the clinical outcomes and intervention duration regarding helmet therapy.
Objectives: :
This study aims to compute an AI-based prediction model on forecasting the intervention duration and clinical outcomes of helmet therapy on positional plagiocephaly by retrospective data, evaluate the accuracy of AI prediction model and improve orthotic helmet therapy service for offering a personalized helmet therapy duration to patients.
Methodology: :
39 completed subjects’ data between 2022 - 2025 in Kowloon Hospital, was input for AI-based model learning. LSTM model was designed under DeepSeek-V3.2 for deep learning. Helmet therapy protocol including monthly measurement and helmet prescription within 14 working days were input in cell state. 15 subjects collected from August to December 2025 was then used for AI prediction with actual result. Both CVAI and head circumference, masked data starting from 2nd month, was input for model prediction. Paired t-test for comparing wind-off time, CVAI and head circumference each month, the CVAI reduction and head circumference growth between AI predicted and actual clinical result were evaluated, as well as the accuracy.
Result & Outcome: :
No significant difference on wind-off time was noted between AI predictor (mean=3.47 months) and the actual result (mean=3.07 months) (p=0.257). Ten out of fifteen subjects’ wind-off time were exactly matched with AI result, with 86.7% accuracy on wind-off time prediction within ±1 month. Borderline significant on CVAI was noted in 2nd (p=0.049) and 3rd month (p=0.054) respectively while no significant difference when comparing the head circumference in all months (p>0.8). AI illustrated a more optimistic trend (rate=-4.41) on CVAI reduction than the actual trend (rate=-3.27) (p< 0.05). Borderline significance was illustrated in head circumference growth rate (p=0.058). Moderate accuracy on CVAI prediction was noted (accuracy=64.6%) while excellent accuracy on circumferential prediction (accuracy=98.2%). Time-to-stability period of additional 2 – 3 follow-up sessions was suggested by AI model. This AI prediction model was suggested as the quick reference with preliminary expectation on intervention duration, thus reassurance can be given to caregivers.
