Advanced

New Methods for Automated Interpretation of Electrocardiograms and Lung Scintigrams

Holst, Holger LU (2000)
Abstract (Swedish)
Popular Abstract in Swedish

Datorer har blivit en viktig del inom den moderna medicinen. Resultat från patientundersökningar såsom bilder, kurvor och mätvärden bearbetas ofta av datorer och presenteras sedan på ett överskådligt sätt för läkaren som grundar sitt svar på bland annat denna information.



Denna doktorsavhandling beskriver hur datorn kan hjälpa till vid själva tolkningen av undersökningsresultaten och föreslå en möjlig diagnos. Detta kallas datorbaserat beslutsstöd. I detta arbete har beslutsstöden det gemensamt att de baseras på en metod som kallas ”artificiella neurala nätverk”. Med denna metod försöker man efterlikna nervsystemets funktion och struktur. Det är särskilt den unika förmågan till... (More)
Popular Abstract in Swedish

Datorer har blivit en viktig del inom den moderna medicinen. Resultat från patientundersökningar såsom bilder, kurvor och mätvärden bearbetas ofta av datorer och presenteras sedan på ett överskådligt sätt för läkaren som grundar sitt svar på bland annat denna information.



Denna doktorsavhandling beskriver hur datorn kan hjälpa till vid själva tolkningen av undersökningsresultaten och föreslå en möjlig diagnos. Detta kallas datorbaserat beslutsstöd. I detta arbete har beslutsstöden det gemensamt att de baseras på en metod som kallas ”artificiella neurala nätverk”. Med denna metod försöker man efterlikna nervsystemets funktion och struktur. Det är särskilt den unika förmågan till mönsterigenkänning som utnyttjas. Artificiella neurala nätverk lär sig genom att träna på patientfall där den korrekta diagnosen är känd.



Doktorsavhandlingen består av fyra delarbeten. I de två första arbetena presenteras metoder som gör att beslutsstödet, förutom att ange om patienten är frisk eller sjuk, även kan ge ett mer nyanserat svar. I svaret kan bland annat ingå uppgifter om beslutsstödet har tidigare erfarenhet av liknande fall och om det diagnostiska problemet var lätt eller svårt. Med den här typen av information kommer läkare att betrakta beslutsstöd som ett mer pålitligt hjälpmedel.



Delarbete tre visar att det är möjligt att utveckla en helautomatisk metod för analys och tolkning av lungbilder. Dessa bilder används för att bedöma om en patient har blodproppar i lungorna.



Det fjärde delarbetet visar att vår helautomatiska metod kan tolka lungbilder från ett annat sjukhus. Ett beslutsstöd riskerar att fungera sämre om det används på sjukhus med annan undersökningsutrustning än den som utnyttjades vid utvecklingen av beslutsstödet. Trots denna skillnad blev beslutsstödets tolkning lika bra som när tre erfarna läkare tolkade samma bilder.



De nya metoderna, som har utvecklats for tolkning av EKG och lungbilder, är generellt tillämpbara och de kan modifieras till att tolka andra typer av patientdata. (Less)
Abstract
The principal purposes of the studies reported in this thesis were to develop new methods to make computerised decision support systems in medicine both more reliable and fully automated. These aims were pursued in the belief that the solution to these problems will lead to an increased clinical use of such systems resulting in benefit to both patient and physician.



In the first two studies, methods to produce measures of confidence for the advice presented by the computer were developed. Although the studies were evaluated by the use of ECGs, the methods developed are generally applicable to other data sources. In the third and fourth studies, a fully automated method for the interpretation of lung scintigrams was... (More)
The principal purposes of the studies reported in this thesis were to develop new methods to make computerised decision support systems in medicine both more reliable and fully automated. These aims were pursued in the belief that the solution to these problems will lead to an increased clinical use of such systems resulting in benefit to both patient and physician.



In the first two studies, methods to produce measures of confidence for the advice presented by the computer were developed. Although the studies were evaluated by the use of ECGs, the methods developed are generally applicable to other data sources. In the third and fourth studies, a fully automated method for the interpretation of lung scintigrams was developed and thereafter evaluated using patient images from a hospital separate from where it was developed. All the studies reported upon were based on the use of artificial neural networks, and supplemented with software for confidence measure calculations or image processing.



The results showed that the methods developed in the first two studies could indicate whether the computer considered a tested ECG to be unusual as well as indicate if it considered the sample to be difficult or easy to interpret. Study three showed that it was possible to develop a fully automated method for image analysis with a performance similar to methods requiring manual assistance by physicians. The fourth study showed that the performance of this automated method was similar to that of experienced physicians when tested on images originating from a different hospital. (Less)
Please use this url to cite or link to this publication:
author
opponent
  • Jorfeldt, Lennart, Professor, Department of Clinical Physiology, Karolinska institutet, Stockholm
organization
publishing date
type
Thesis
publication status
published
subject
keywords
Decision Support System (Clinical), Image Processing (Computer-Assisted), Radionucleotide Imaging, Pulmonary Embolism, tomography, radiology, Neural Networks (Computer), Electrocardiography, Clinical physics, medical instrumentation, Klinisk fysiologi, radiologi, tomografi, medicinsk instrumentering
pages
100 pages
publisher
Department of Clinical Physiology, Lund University
defense location
F3 at the University Hospital in Lund
defense date
2000-11-27 13:00
external identifiers
  • Other:ISRN: LUMEDW/MEFL--1018--SE
language
English
LU publication?
yes
id
d04e74f7-111c-44f0-a77c-6622a9d47134 (old id 41008)
date added to LUP
2007-06-21 09:46:22
date last changed
2016-09-19 08:45:07
@misc{d04e74f7-111c-44f0-a77c-6622a9d47134,
  abstract     = {The principal purposes of the studies reported in this thesis were to develop new methods to make computerised decision support systems in medicine both more reliable and fully automated. These aims were pursued in the belief that the solution to these problems will lead to an increased clinical use of such systems resulting in benefit to both patient and physician.<br/><br>
<br/><br>
In the first two studies, methods to produce measures of confidence for the advice presented by the computer were developed. Although the studies were evaluated by the use of ECGs, the methods developed are generally applicable to other data sources. In the third and fourth studies, a fully automated method for the interpretation of lung scintigrams was developed and thereafter evaluated using patient images from a hospital separate from where it was developed. All the studies reported upon were based on the use of artificial neural networks, and supplemented with software for confidence measure calculations or image processing.<br/><br>
<br/><br>
The results showed that the methods developed in the first two studies could indicate whether the computer considered a tested ECG to be unusual as well as indicate if it considered the sample to be difficult or easy to interpret. Study three showed that it was possible to develop a fully automated method for image analysis with a performance similar to methods requiring manual assistance by physicians. The fourth study showed that the performance of this automated method was similar to that of experienced physicians when tested on images originating from a different hospital.},
  author       = {Holst, Holger},
  keyword      = {Decision Support System (Clinical),Image Processing (Computer-Assisted),Radionucleotide Imaging,Pulmonary Embolism,tomography,radiology,Neural Networks (Computer),Electrocardiography,Clinical physics,medical instrumentation,Klinisk fysiologi,radiologi,tomografi,medicinsk instrumentering},
  language     = {eng},
  pages        = {100},
  publisher    = {ARRAY(0x94794e8)},
  title        = {New Methods for Automated Interpretation of Electrocardiograms and Lung Scintigrams},
  year         = {2000},
}