Three-component IVIM fitting in cerebrovascular disease using physics-informed neural networks: repeatability and accuracy
Paulien H.M. Voorter1,2, Walter H. Backes1,2,3, Oliver J. Gurney-Champion4, Sau-May Wong1, Julie Staals3,5, Robert-Jan van Oostenbrugge2,3,5, Merel M. van der Thiel1,2, Jacobus F.A. Jansen1,2,6, and Gerhard S. Drenthen1,2
1Department of Radiology & Nuclear Medicine, Maastricht University Medical Center, Maastricht, Netherlands, 2School for Mental Health & Neuroscience, Maastricht University, Maastricht, Netherlands, 3School for Cardiovascular Disease, Maastricht University, Maastricht, Netherlands, 4Department of Radiology and Nuclear medicine, Amsterdam University Medical Center, Cancer Center Amsterdam, University of Amsterdam, Amsterdam, Netherlands, 5Department of Neurology, Maastricht University Medical Center, Maastricht, Netherlands, 6Department of Electrical Engineering, Eindhoven University of Technology, Eindhoven, Netherlands
Next to parenchymal diffusion and microvascular pseudo-diffusion, a third diffusion component is present in cerebral intravoxel incoherent motion (IVIM) imaging, representing interstitial fluid. Fitting the three-component IVIM model using conventional fitting methods strongly suffers from image noise. Therefore, we explored the applicability of a physics-informed neural network (PI-NN) fitting approach, previously shown to be more robust to noise. Using test-retest data from sixteen patients with cerebrovascular disease, we found higher repeatability of all IVIM parameters using PI-NN. Furthermore, simulations showed that PI-NN provided more accurate IVIM parameters. Hence, using PI-NN is promising to obtain tissue markers of cerebrovascular disease.
This abstract and the presentation materials are available to members only;
a login is required.