CCP PET-MR

Collaborative Computational Project in Positron Emission Tomography
and Magnetic Resonance imaging

Professor Johan Nuyts: Reconstruction with MR-prior for PET brain imaging

BMEIS Seminar Series

The School of Biomedical Engineering and Imaging Sciences Seminar Series

Mon 10th Dec 2018, 12.00 to 13.00

The Large Seminar Room, 4th Floor Lambeth Wing,

St Thomas’ Hospital.

https://ukri-stfc.zoom.us/j/364764386   

 

Professor Johan Nuyts

 

“Reconstruction with MR-prior for PET brain imaging”

 

Bio:

Johan Nuyts is professor of the Faculty of Medicine at KULeuven, Leuven Belgium. He is with the Department of Nuclear Medicine and with the Medical Imaging Research Center (MIRC), and is Honorary Professor in Medical Radiation Sciences, Faculty of Health Sciences, University of Sydney. He co-authored about 150 scientific journal papers and 5 patents. His research focus is iterative reconstruction in PET, SPECT and CT. Ongoing research projects focus on maximum-a-posteriori reconstruction in emission tomography, iterative reconstruction in CT, correction for attenuation in PET/CT, PET/MRI and TOF-PET and multi-modal imaging for Selective Internal Radiation Therapy (SIRT).

 

 

Abstract:

Since many years, researchers have attempted to improve PET image quality by exploiting anatomical side-information obtained from other modalities. PET data suffer from limited spatial resolution and high noise levels. By modelling the resolution in the system matrix, iterative reconstruction algorithms will attempt to compensate for it. However, resolution recovery as an ill-posed problem and unconstrained application often leads to disturbing high frequency artefacts. To suppress these artefacts and reduce the noise without losing the recovered resolution, the image reconstruction process can be guided with high resolution anatomical images. This approach is rather effective in PET/MR brain imaging, because MR provides highly detailed anatomical images, and for many tracers, there is a strong correlation between tracer uptake and anatomy. The introduction of hybrid PET/MR imaging facilitates such multimodal image processing approaches.

Several algorithms have been proposed for PET reconstruction with anatomical priors, including different flavours of the parallel level-sets prior and the Bowsher prior. Preliminary evaluation studies indicate that these priors improve quantitative accuracy, benefit lesion detection by human observers and may enable dose reduction. Nevertheless, the algorithms have not yet been introduced in clinical practice, probably because the resulting PET images look very different from the conventional ones and because the dedicated maximum-a-posterior reconstruction is complicated and slow. In an attempt to mitigate this second issue, we are training a convolutional neural net to predict the Bowser reconstruction from a conventional MLEM (OSEM) reconstruction (with resolution recovery) and the aligned MR-image. Current results show that if the network is trained with a sufficiently diverse set of images, it performs well, even on PET images with tracers not included in the training set. When run on a state-of-the-art GPU, the prediction time of the trained network is less than a second. We hope that this will facilitate introduction of these methods in clinical practice, which in turn will enable a systematic evaluation of the method in different PET brain imaging applications.

Directions to The Large Seminar room, St Thomas’ Hospital, SE1 7EH

*The nearest tube stations are Waterloo & Westminster, both a 5 to 10 minute walk away. Please enter the hospital via the main entrance off Westminster Bridge Road and turn left onto the Lambeth Wing. Follow the corridor all the way round until you come to the lifts and take the lifts to the 4th floor. Upon exiting the Large Seminar Room is directly opposite the lifts just to the left.

You'd be quite welcome to attend physically or remotely via the following Zoom link: 

https://ukri-stfc.zoom.us/j/364764386   

 

Date: 
Monday, December 10, 2018 - 12:00 to 13:00