Our main reserach focus is positron emission tomography (PET) and other medical imaging techniques. Especially, we investigate PET image reconstruction methods, including attenuation and motion correction. We also develop image processing tools for medical purposes, includig machine learning methods.

Medical tomography image analysis using AI

Applications of artificial intelligence (AI) and machine learning (ML) has revolutionalised image analysis. Examples of ML applications in image analysis include autonumous cars and face recognition. We apply these ML methods to medical tomography image analysis concentrating on PET.

Motion correction of PET images

PET image acquisitions take several minutes and a subject motion may occur during the acquisition. Correcting of the motion increases the quality of the PET images. We research motion correction methods for cardiac and brain imaging deploying data-driven methods and external sensors.

MRI based attenuation correction of PET images

Accurate attenuation correction for PET/MR imaging is essential for accurate quantification and visual analysis of PET images. We investigate new methods to improve MR-based attenuation correction for neurological imaging.

Image quantification accuracy of O-15 H2O myocardial perfusion imaging

With the increasing number of O-15 H2O PET myocardial perfusion scans, it is important to investigate the factors affecting image quantification accuracy and the reproducibility of PET systems in perfusion imaging. This allows to minimise inaccuracies and uncertainties in perfusion measurements. Part of this work is done in collaboration with the National Physical Laboratory (United Kingdom) in a joint project TracPETperf (

New kinetic modelling approaches for total body PET

Recent advances in PET instrumentation have enabled to construct total body PET scanners with axial FOV of over 100 cm, allowing imaging of the entire body and it’s physiology simultaneously. The typical modelling approach of PET images is based on applying compartmental models, which are optimal for single-organ imaging. However, in total body PET with full organ coverage and interaction, new modelling approaches may be required. We will investigate new modelling methods appliable for total body PET imaging. For more details see: (or in Finnish FIN: