% IMPORTANT: The following is UTF-8 encoded. This means that in the presence
% of non-ASCII characters, it will not work with BibTeX 0.99 or older.
% Instead, you should use an up-to-date BibTeX implementation like “bibtex8” or
% “biber”.
@PHDTHESIS{Lyra:1012152,
author = {Lyra, Simon Alexander},
othercontributors = {Leonhardt, Steffen and Joseph, Jayaraj and Orlikowsky,
Thorsten},
title = {{C}amera-based vital signs monitoring of neonates in
real-time using deep learning},
volume = {79},
school = {RWTH Aachen University},
type = {Dissertation},
address = {Düren},
publisher = {Shaker},
reportid = {RWTH-2025-04894},
isbn = {978-3-8191-0119-9},
series = {Aachener Beiträge zur Medizintechnik},
pages = {XXI, 209 Seiten : Illustrationen},
year = {2025},
note = {Dissertation, RWTH Aachen University, 2025},
abstract = {Almost one in ten children worldwide is born prematurely.
Due to their immature physiology, these newborns require
extensive medical care and constant monitoring to support
the development process and minimize associated
complications. Today’s state-of-the-art monitoring
techniques include wired sensors and electrodes attached to
the patient’s bare skin with medical-grade adhesives.
While close monitoring of vital signs is a key factor in
neonatal care to intervene in pathological progression, the
adhesives pose a significant risk to the immature and
therefore vulnerable skin, as they can cause injuries that
can lead to life-threatening infections. Thus, alternative
techniques for continuous monitoring of neonatal patients
that allow unobtrusive remote assessment of vital signs have
the potential to replace wired sensors and improve the
overall outcome of neonatal care by minimizing the risk of
skin injury. One technology that is increasingly being
investigated to meet these requirements is camera-based
sensors that can measure vital signs in real-time without
direct skin contact. This thesis presents novel approaches
to multi-modal camera-based monitoring of vital signs in
preterm infants with the aim of improving neonatal care.
Three different software pipelines have been implemented
using real-time image processing techniques from the rapidly
advancing and innovative field of Deep Learning. These tools
were validated by analyzing multi-modal image data from
neonatal patients acquired during two clinical studies and
one laboratory study using a novel phantom implemented with
hardware components to simulate vital signs. The two
clinical trials recorded a total of 33 subjects in neonatal
intensive care units in India and Germany. The data was
processed to extract cardiac activity, respiration,
thermoregulation, and patient movement. Each study was
conducted using a specially designed camera system,
including an RGB camera and Infrared Thermography equipment
for thermal imaging. The neonatal phantom was manufactured
using rapid prototyping and used to validate a low-cost
camera setup by simulating pathological conditions of
cardiac and thermoregulatory processes monitored in
real-time. To address the challenge of the limited
availability of public medical image data, which is crucial
for training data-driven prediction models, a novel approach
for Deep Learning-based data augmentation was applied to the
clinic-recorded data. The results presented demonstrate the
potential of Deep Learning-based image processing for
real-time monitoring of vital signs in neonates and show the
possibilities of using hardware phantoms for laboratory
studies, reducing expensive and time-consuming clinical
trials. Finally, this work represents a step towards
replacing potentially harmful wired sensors in neonatal
care.},
cin = {611010},
ddc = {621.3},
cid = {$I:(DE-82)611010_20140620$},
typ = {PUB:(DE-HGF)11 / PUB:(DE-HGF)3},
url = {https://publications.rwth-aachen.de/record/1012152},
}
guest ::