Use of artificial intelligence to enhance the analysis of myocardial perfusion PET studies

Positron emission tomography (PET) imaging provides reliable data for the detection of cardiovascular diseases, and diagnostics can be further enhanced by using artificial intelligence. Our research team is currently developing methods based on artificial intelligence that would enable automated analyses of myocardial perfusion PET studies. This would make the interpretation of the PET results easier.

Published 11.1.2024
Text: Jarmo Teuho
Image: Shutterstock
Editing: Viestintätoimisto Jokiranta Oy


Positron emission tomography (PET) imaging can be used for detecting cardiovascular diseases. Myocardial perfusion imaging (MPI) with PET provides information about the risk of severe cardiac events, thereby guiding the choice of treatment. The reliability of MPI PET has been evaluated in several studies, and it has been found to be the most accurate of the available non-invasive methods. For the diagnostics of coronary artery disease (CAD), in particular, the quantification of cardiac blood flow is a major advantage of PET studies.

Currently, perfusion imaging analyses are conducted using Carimas, an image analysis software developed specifically for this purpose at the Turku PET Centre in Finland. Our research team is examining the development of methods based on artificial intelligence (AI) and their application in the various areas of MPI analyses to enable the evolution and automation of the analyses. Our work involves research in the fields of medicine, imaging and engineering.


International research collaboration

Artificial intelligence and, in particular, deep learning is already efficiently used in, for example, pattern recognition, image processing, data analytics and, most recently, in many types of content production. Of the medical applications, the best known are probably certain tools for COVID diagnostics, whose development was extensively covered by the media. Moreover, AI is already rather widely used in other fields of medical imaging, such as CT and MRI.

In MPI PET studies, however, methods based on AI have not yet been applied to any large extent. Our work on the application of AI methods in PET imaging started in 2019 through collaboration with a research team in Japan that specialised in machine learning. This collaboration focuses on the study of deep learning methods and their application in PET imaging. Recently, during my 12-month period as a visiting researcher at the Nara Institute of Science and Technology in Japan, I had an opportunity to develop new PET imaging data analysis methods. Our current research is a direct continuation of this collaboration.


Useful tools for diagnostics

The application of AI is useful in PET imaging and diagnostics and, in particular, for the automation of the analysis chain. AI models can be applied, for example, to the automated definition of areas of interest (image segmentation), the modelling of physiological parameters, such as blood flow (data modelling) and the automated diagnostics based on different data (results classification). These methods will make clinical work faster and more efficient, while also improving the repeatability of analysis.

The current models can be combined with techniques that enable us to explain the model predictions and behaviours in a comprehensible manner. This helps in the interpretation of the AI model – unlike the "black box” model, where the decision-making chain and motivations remain vague.

In addition to methodological development, one aim of the project is to ensure the developed models are easy to interpret and explain. Thus, the outcome generated by AI is always verifiable and interpretable by users, which is a vital aspect in diagnostic studies.


AI in the automation of analysis

In our current project, we are developing deep-learning models for the automation of the PET perfusion analysis chain. Models are being developed for image segmentation, data modelling and classification. Moreover, we will compare the outcomes of modelling with the results of clinical analyses in order to assess the accuracy and functioning of the models. This will make the entire analysis chain faster, more repeatable and more accurate. Our aim is to integrate the tools being generated in the project with the Carimas software, originally developed at Turku PET Centre.

We are grateful to the Sakari Alhopuro Foundation for awarding the grant that allowed us to implement our research to this extent and on this schedule. Thank you for supporting science and research!


Jarmo Teuho.



Jarmo Teuho, Doctor of Medical Physics and Engineering, is working as a researcher at the Turku PET Centre, Finland. In 2018, he earned his doctoral degree with a dissertation concerning PET/MR imaging and, in 2021, he was awarded the title of Adjunct Professor (title of Docent) in Medical Imaging Physics and Technology at the University of Turku. Teuho has worked in Japan as a visiting scholar twice, at the Nara Institute of Science and Technology in Nara and the National Cerebral and Cardiovascular Research Centre in Osaka. Currently, he and his team are engaged in research concerning technical method development for PET imaging, including machine learning methods.





References (in Finnish):

Read more