Signal Quality in Continuous Transcutaneous Bilirubinometry

Area of Science:

• Biomedical Engineering
• Medical Diagnostics
• Spectroscopy

Background:

• Bilirubin monitoring is crucial for newborns due to jaundice risks.
• Current methods (visual, transcutaneous bilirubinometry) have limitations in accuracy across skin tones and reliability.
• A continuous monitoring device is under development to improve neonatal jaundice detection.

Purpose of the Study:

• To develop a machine learning model for predicting the placement status of a continuous bilirubin monitoring device.
• To establish a signal quality indication index for reliable bilirubin concentration estimates.
• To validate the model's performance on adult skin spectra from different body areas.

Main Methods:

• Utilized machine learning (ML), specifically Support Vector Machine (SVM) models.
• Acquired skin spectra measurements from adult participants.
• Trained and tested SVM models on spectral data from arm and forehead regions.

Main Results:

• SVM models demonstrated high performance in predicting device placement status.
• The developed method serves as an intermediate step for ensuring reliable bilirubin measurements.
• Results indicate the potential for improving continuous monitoring device accuracy across diverse skin tones.

Conclusions:

• Machine learning, particularly SVM, is effective for predicting optimal device placement for bilirubin monitoring.
• This approach enhances signal quality assessment, crucial for accurate neonatal jaundice detection.
• The findings support the development of more reliable, continuous, and non-invasive bilirubin monitoring systems.