A Novel Artificial Intelligence-Based Mobile Application for Pediatric Weight Estimation

Background/Objectives: Pediatric drug dosages are typically weight-based. Length-based weight estimation tools used in emergency situations require full body extension, which may cause measurement errors in restricted positions. In this study, we developed and evaluated a weight prediction application using MoveNet's human pose estimation and a deep neural network (DNN) regression model. Methods: This prospective cross-sectional study was conducted from June 2023 to May 2024 and included pediatric patients aged 1 month to 12 years. Weight estimation accuracy was compared between the Pediatric Artificial Intelligence weight-estimating Camera (PAICam) and the Broselow tape (BT) using mean percentage error (MPE), mean absolute percentage error (MAPE), and root mean square percentage error (RMSPE). The percentages of weight estimations within 10% (PW10) and 20% (PW20) of the actual weights were calculated. Intraclass correlation coefficients (ICCs) were used to evaluate agreement between predicted and actual weights. Results: In total, 1335 pediatric participants were analyzed (57.4% boys, 42.6% girls), with an average age of 4 years. The BT and PAICam showed comparable performance, with similar values for MPE (-1.44% vs. 5.29%), MAPE (11.28% vs. 12.41%), and RMSPE (3.09% vs. 3.42%). PW10 and PW20 for the BT and PAICam were also similar (52.6% vs. 51.2% and 79.1% vs. 77.7%). ICC values demonstrated strong agreement between actual and predicted weights for both methods (0.959 vs. 0.955). Conclusions: PAICam, utilizing deep learning and human pose estimation technology, demonstrated performance and accuracy comparable to the BT. This suggests its potential as an alternative tool for pediatric weight estimation in emergency settings.