Machine learning assisted noncontact neonatal anthropometry using FMCW radar

Area of Science:

• Biomedical Engineering
• Neonatal Care
• Machine Learning

Background:

• Accurate measurement of neonatal anthropometrics is crucial for monitoring growth and health.
• Traditional methods can be cumbersome and may cause distress to newborns.
• Advancements in sensor technology and artificial intelligence offer potential for non-invasive monitoring solutions.

Purpose of the Study:

• To develop and validate a method for convenient, safe, and accurate measurement of neonatal height and weight.
• To apply a convolutional neural network (CNN) to frequency-modulated continuous-wave (FMCW) radar data for anthropometric assessment.
• To evaluate the performance of the CNN-based FMCW radar system in measuring neonatal physical parameters.

Main Methods:

• Utilized frequency-modulated continuous-wave (FMCW) radar sensors to collect data from neonates.
• Applied a convolutional neural network (CNN) model for processing radar data and estimating height and weight.
• Trained and tested the CNN model on data from fifteen neonates, divided into training/testing and separate testing groups.

Main Results:

• The model achieved high accuracy in height measurements with a mean absolute error (MAE) of 1.34 cm and intraclass correlation coefficient (ICC) of 0.78 in the training group.
• Weight measurements demonstrated an MAE of 0.23 kg and an ICC of 0.85 in the training group.
• Validation on a separate group showed comparable results, with MAE of 1.51 cm for height and 0.20 kg for weight, and ICCs of 0.68 and 0.75, respectively.

Conclusions:

• FMCW radar combined with CNN provides a feasible and accurate method for non-invasive neonatal anthropometric measurements.
• This technology has the potential for continuous monitoring of neonatal growth and health status.
• The proposed method offers a convenient and practical alternative to traditional measurement techniques in neonatal care.