Advanced

Binary classification of HRV signals

Lütz, Rebecca (2019) MASK01 20191
Mathematical Statistics
Abstract
Heart rate variability, commonly abbreviated as HRV, displays the variance
between consecutive heartbeats. This variance occurs naturally but can change
due to stress and problems with the cardiac system. HRV is therefore widely
used for medical research. The goal of this thesis is to correctly classify two
HRV signals where one is obtained at a resting state, the warm signal, while
the cold signal is obtained during a simulation of stress. The use of spectral
estimation methods leads to the analysis of the high frequency range (0.12 -
0.4 Hz) as well as the analysis of a more narrow frequency band around the
respiratory maximum. The analysis of those frequency ranges is done by using
linear models as well as studying how the... (More)
Heart rate variability, commonly abbreviated as HRV, displays the variance
between consecutive heartbeats. This variance occurs naturally but can change
due to stress and problems with the cardiac system. HRV is therefore widely
used for medical research. The goal of this thesis is to correctly classify two
HRV signals where one is obtained at a resting state, the warm signal, while
the cold signal is obtained during a simulation of stress. The use of spectral
estimation methods leads to the analysis of the high frequency range (0.12 -
0.4 Hz) as well as the analysis of a more narrow frequency band around the
respiratory maximum. The analysis of those frequency ranges is done by using
linear models as well as studying how the energy of the cold and the warm signal
is distributed. All approaches lead to binary classification with more than 50%
accuracy. However, the best results are obtained when analyzing the frequency
band around the respiratory maximum located at 0.2 Hz or higher. When using
a linear model for changes in energy over time, dividing the data into four sets
leads to 93.4% correct classification. When analyzing the energy that is present
in the first 90 s of each signal, 96.23% correct classification is obtained. (Less)
Popular Abstract (Swedish)
HRV är en förkortning av det engelska begreppet ”Heart rate variability” och
beräknas med hjälp av data fr˚an ett elektrokardiogram, EKG. Detta mäter elek-
triska signaler som skickas ut fr˚an hjärtat. En tydlig sammandragning av hjärtat
kallas för puls och det syns tydlig p˚a ett EKG som en ökning i den elektriska
signalen. I motsats till pulsen, är det inte själva hjärtslagen som är viktiga
när man beräknar HRV utan det som händer emellan dem. HRV kan därför
definieras som variansen av den elektriska signalen fr˚an hjärtat mellan tv˚a ty-
dliga sammandragningar som vi kallar puls. Dessa intervaller mellan hjärtslagen
är inte identiska vilket leder till en naturlig varians. Men stora avvikelser inom
den naturliga variansen kan... (More)
HRV är en förkortning av det engelska begreppet ”Heart rate variability” och
beräknas med hjälp av data fr˚an ett elektrokardiogram, EKG. Detta mäter elek-
triska signaler som skickas ut fr˚an hjärtat. En tydlig sammandragning av hjärtat
kallas för puls och det syns tydlig p˚a ett EKG som en ökning i den elektriska
signalen. I motsats till pulsen, är det inte själva hjärtslagen som är viktiga
när man beräknar HRV utan det som händer emellan dem. HRV kan därför
definieras som variansen av den elektriska signalen fr˚an hjärtat mellan tv˚a ty-
dliga sammandragningar som vi kallar puls. Dessa intervaller mellan hjärtslagen
är inte identiska vilket leder till en naturlig varians. Men stora avvikelser inom
den naturliga variansen kan tyda p˚a att n˚agonting s˚a som stress eller sjukdom
p˚averkar kroppen. Till exempel leder stress till att variansen minskar.
Datan som har använts i denna uppsats är tagen fr˚an en studie i Kristanstad
där 53 personer placerade sin hand i iskallt vatten. Detta skulle simulera stress
och visa hur stress p˚averkar HRV. För att f˚a ett kontrollset utfördes en liknande
test med varmt vatten vilket visade HRV i viloläget.
Denna uppsats behandlar binära klassificeringar vilket innebär att data blir
antingen klassificerad som positiv eller negativ. I v˚art fall betyder en positiv
klassificiering att signalen har blivit klassificierad som kall. Den negativa klas-
sificieringen betyder s˚aledes att signalen har klassificierats som varm. M˚alet
med denna uppsats är att f˚a fram en metod för lyckad klassificiering som är
baserad p˚a en frekvensanalys. Detta betyder att man analyserar HRV signalen
med fokus p˚a vilka frekvenser som finns och hur mycket energi som har acku-
mulerats vid dessa frekvenser.
Medan olika metoder leder till olika bra resultat är det änd˚a väldigt tydligt
att det verkar finnas parametrar utifr˚an vilka man kan korrekt klassificiera tv˚a
okända HRV signaler. Denna slutsatsen kan dras eftersom alla metoder klassi-
ficierar minst 50% av datan rätt. (Less)
Please use this url to cite or link to this publication:
author
Lütz, Rebecca
supervisor
organization
course
MASK01 20191
year
type
M2 - Bachelor Degree
subject
language
English
id
8986844
date added to LUP
2019-06-20 17:11:53
date last changed
2019-06-20 17:11:53
@misc{8986844,
  abstract     = {Heart rate variability, commonly abbreviated as HRV, displays the variance
between consecutive heartbeats. This variance occurs naturally but can change
due to stress and problems with the cardiac system. HRV is therefore widely
used for medical research. The goal of this thesis is to correctly classify two
HRV signals where one is obtained at a resting state, the warm signal, while
the cold signal is obtained during a simulation of stress. The use of spectral
estimation methods leads to the analysis of the high frequency range (0.12 -
0.4 Hz) as well as the analysis of a more narrow frequency band around the
respiratory maximum. The analysis of those frequency ranges is done by using
linear models as well as studying how the energy of the cold and the warm signal
is distributed. All approaches lead to binary classification with more than 50%
accuracy. However, the best results are obtained when analyzing the frequency
band around the respiratory maximum located at 0.2 Hz or higher. When using
a linear model for changes in energy over time, dividing the data into four sets
leads to 93.4% correct classification. When analyzing the energy that is present
in the first 90 s of each signal, 96.23% correct classification is obtained.},
  author       = {Lütz, Rebecca},
  language     = {eng},
  note         = {Student Paper},
  title        = {Binary classification of HRV signals},
  year         = {2019},
}