Artificial neural networks classify myocardial perfusion images

Dan Lindahl

Research output: ThesisDoctoral Thesis (compilation)

Abstract

In the studies of this thesis, a method for automated classification of myocardial perfusion images was successfully developed and evaluated. The results show that the method, based on artificial neural networks, was equally good, or even better than human experts. It was also found that physicians interpreting myocardial perfusion images benefit from the advice of the artificial neural networks. Furthermore the neural networks could be trained to present clinical interpretations including the information regarding extent and severity of reversible and irreversible defects. At last, it was shown that the networks can maintain it's high accuracy also in a hospital separate from where it was developed. In conclusion, these studies show the feasibility of developing a decision-support system for the interpretation of myocardial perfusion images. The clinical implication of such a decision-support system could be significant. For patients newly presenting with possible coronary artery disease, it has been shown that investigative strategies using myocardial perfusion imaging are more cost-effective than strategies that don't. Decision support for the interpretation of myocardial perfusion images, such as artificial neural network, could further improve this cost-effectiveness.
Original languageEnglish
QualificationDoctor
Awarding Institution
  • Clinical Physiology (Lund)
Supervisors/Advisors
  • [unknown], [unknown], Supervisor, External person
Award date2000 May 23
Publisher
Publication statusPublished - 2000

Bibliographical note

Defence details

Date: 2000-05-23
Time: 14:00
Place: Föreläsningssal 1, Centralblocket, Universitetssjukhuset i Lund

External reviewer(s)

Name: Carlens, Per
Title: Docent
Affiliation: Fysiologiska kliniken, Södersjukhuset, Stockholm

---

Subject classification (UKÄ)

  • Cardiac and Cardiovascular Systems
  • Respiratory Medicine and Allergy

Free keywords

  • tomography
  • radiology
  • Clinical physics
  • radionuclide imaging
  • ischemic heart disease
  • computer-assisted diagnosis
  • medical instrumentation
  • Klinisk fysiologi
  • radiologi
  • tomografi
  • medicinsk instrumentering

Fingerprint

Dive into the research topics of 'Artificial neural networks classify myocardial perfusion images'. Together they form a unique fingerprint.

Cite this