← Back to Sahar Yousefi

Medical Imaging · Arterial Spin Labeling

Fast Dynamic Perfusion and Angiography Reconstruction

Read the paper

An end-to-end 3D convolutional network for reconstructing dynamic perfusion and angiographic images from Hadamard time-encoded ASL data.

Authors

Sahar Yousefi, Lydiane Hirschler, Merlijn van der Plas, Mohamed S. Elmahdy, Hessam Sokooti, Matthias J. P. van Osch, and Marius Staring

Overview

Rank-8 Hadamard time-encoded pseudo-continuous arterial spin labeling (te-pCASL) captures temporal information in the labeling sequence. Acquisitions with and without flow crushing can provide both perfusion information and 4D angiographic information. This work investigates reconstructing both from a single interleaved acquisition, reducing the need for separate scans.

Reconstruction approach

The method uses a 3D Dense U-Net with a multi-level loss function to reconstruct multi-timepoint perfusion and angiographic data from interleaved, 50%-sampled flow-crushed and non-crushed measurements. A data-generation framework based on intravascular and extravascular te-pCASL signal models was used to create training and validation data. The paper evaluates the method on 313 test datasets.