Related Experiment Video
Updated: Nov 15, 2025
06:27
Transcutaneous Microcirculatory Imaging in Preterm Neonates
Published on: December 31, 2015
8.4K
Preterm Newborn Presence Detection in Incubator and Open Bed Using Deep Transfer Learning
IEEE Journal of Biomedical and Health Informatics
|March 1, 2021
Summary
This study introduces an automated method to detect preterm newborn presence in incubators and open beds using video analysis. This ensures accurate, non-invasive monitoring by excluding periods without the infant or with adult presence.
Area of Science:
- Neonatal intensive care
- Medical imaging
- Machine learning
Background:
- Video-based motion analysis offers non-invasive monitoring for preterm newborns in NICUs.
- Accurate analysis requires excluding periods of newborn absence or adult presence.
Purpose of the Study:
- To develop an automated method for detecting preterm newborn presence in both incubator and open bed settings.
- To improve the reliability of contactless, non-invasive monitoring systems.
Main Methods:
- A deep transfer learning approach using fused binary classifiers for newborn and adult presence detection.
- Separate models were trained for incubator and open bed environments due to differing camera placements and situations.
- Utilized pre-trained deep neural networks (VGG16, MobileNetV2, InceptionV3) and compared Support Vector Machine (SVM) with a small neural network classifier.
- Employed decision interval fusion for enhanced temporal consistency.
Main Results:
- Achieved 86% balanced accuracy for newborn presence detection in incubators.
- Performance in open beds was lower due to greater environmental variability and limited data.
Conclusions:
- The proposed method demonstrates effectiveness in automated newborn presence detection, particularly in incubator settings.
- Further refinement is needed for open bed scenarios to address data limitations and environmental diversity.
