The program will be available shortly. Please check back later.
Presenters: Samuel Kuttner1,2,3, Luigi Luppino2
1PET Imaging Center, University Hospital of North Norway, Tromsø.
2Machine Learning Group, Department of Physics and Technology, UiT The Arctic University of Norway, Tromsø, Norway.
3Nuclear Medicine and Radiation Biology Research Group, Department of Clinical Medicine, UiT The Arctic University of Norway, Tromsø, Norway.
Abstract:
Positron emission tomography (PET) is a medical imaging technique used to visualize biological processes in vivo and non-invasively. It is widely used in human clinical imaging diagnosis and preclinical PET imaging applied to small animals. Dynamic PET imaging using various PET tracers allows for measuring biological processes, e.g., glucose metabolism, receptor binding, or up-regulation of amino acid transporters. This requires tissue uptake curves from PET imaging and an arterial input function (AIF), from blood sampling. Arterial cannulation in humans is, however, laborious, time-consuming, and potentially painful, with a risk for complications. Also, in small animals, arterial blood sampling is hampered by limited blood volume and complex and terminal surgery for arterial cannulation. Several methods have been proposed to overcome these limitations, e.g., population-based AIF, image-derived input function, or simultaneous estimation. However, most of these methods have limited usability because they still require a blood sample for calibration, lack the necessary precision, or have limitations related to the spatial and temporal resolution of the PET scanner.
In the current work, we aim to develop and evaluate a novel, non-invasive, and fully automatic deep learning-derived input-function (DLIF) model to estimate the AIF from dynamic PET imaging. To establish a training database for the DLIF model, we investigate how different variables affect the AIF, such as tracer injection volume, injection time, sampling rate, mouse age. For model prediction, we evaluate three and four-dimensional convolutional neural networks that we have adopted for medical image time-series regression.
The project is part of an innovation track with potential for commercialization. We will discuss efforts and opportunities that the collaboration with the Technology Transfer Office (Norinnova) has added to the project.
In compliance with GDPR consent requirements, presentations given in a Visual Intelligence context may be recorded with the consent of the speaker. All recordings are edited to remove all faces, names and voices of other participants. Questions and comments by the audience will hence be removed and will not appear in the recording. With the freely given consent from the speaker, recorded presentation may be posted on the Visual Intelligence YouTube channel.