Machine learning assisted noncontact neonatal anthropometry using FMCW radar

Abstract

This study proposes a method for measuring the height and weight of a neonate conveniently, safely, and accurately by applying a convolutional neural network to frequency-modulated continuous-wave (FMCW) radar sensor data. Fifteen neonates, with parental consent, were enrolled in the study. The neonates were divided into two groups for analysis. Group 1, comprising ten neonates, was used for training and testing the machine learning model. The model achieved a mean absolute error (MAE) of 1.34 cm, a root mean square error (RMSE) of 1.55 cm, and an intraclass correlation coefficient (ICC) of 0.78 (p value < 0.001) in height measurements and an MAE of 0.23 kg, RMSE of 0.28 kg, and ICC of 0.85 (p value < 0.001) in weight measurements. Group 2 comprised five neonates and was used only to test the trained model. The model showed an MAE of 1.51 cm, RMSE of 1.70 cm, and ICC of 0.68 (p value < 0.001) in height measurements, alongside an MAE of 0.20 kg, RMSE of 0.25 kg, and ICC of 0.75 (p value < 0.001) in weight measurements. These results highlight the importance of FMCW radar-based measurements for continuous monitoring of neonatal growth and health, offering a convenient and practical solution.