Improved Interpretation of Myocardial Perfusion Images by Artificial Neural Networks

Sammanfattning: Correct interpretation of medical imaging is based upon the interpreter’s experience and image quality. Depending on the report given the clinician decides how to use the result of the study. For many years a large number of image technologies including myocardial perfusion scintigraphy (MPS) have been acquired in digital format and the development of the imaging modalities are progressing rapidly. The need for software tools facilitating correct interpretation and assuring quality therefore also grows. In this thesis different ways were studied when and where a decision support system (DSS) can be used to improve the interpretation of MPS, mainly looking at two issues: a) The quality of the DSS. We studied how the performance of the DSS is influenced by the size of databases in relation to tracers used and possible image differences between genders (paper I). In paper III we studied the use of artificial neural networks for quality assurance of routine image interpretations. b) The influence of a DSS on the performance of interpreters with different experience levels (paper II) using both MPS stress and rest images. In paper IV we examined how often a rest study could be skipped to save resources and radiation. The results of paper I showed only a minor importance of the database being constructed with patients investigated using the same or different gender or tracers respectively. A large mixed database might even be superior to a smaller tracer-specific database. Some influence was seen for gender specific vs. mixed female-male databases regarding reversible defects and size of databases. We found that a DSS could be used to increase the quality of a DSS in an efficient way (paper III). In paper II it was demonstrated that the DSS improved physicians’ interpretations of reversible defects from a combined stress-rest MPS. For less experienced physicians DSS could make a difference. In paper IV stress-only MPS interpretation regarding the need of getting a rest image was improved. The conclusions in this thesis including the value of a DSS as a diagnostic advice are obviously not limited for MPS interpretation: DSS can be applied to many other types of medical images. The use of tools for supporting the interpretation of medical images will undoubtedly increase in the near future in relation to the rapid development of the imaging modalities.