Original title: Multi-delay arterial spin-labeled perfusion estimation with biophysics simulation and deep learning
Authors: Renjiu Hu, Qihao Zhang, Pascal Spincemaille, Thanh D. Nguyen, Yi Wang
The study aimed to estimate perfusion levels using arterial spin labeling (ASL) images through deep learning. They trained a 3D U-Net named QTMnet to estimate perfusion from 4D tracer propagation images. Testing on simulated tracer concentration data and a synthetic brain ASL image from MR angiography showed QTMnet accurately reconstructed perfusion levels. Comparing its estimations with conventional models in images from eight healthy volunteers, QTMnet demonstrated a significantly lower relative error (7.04%) for perfusion estimation compared to single-delay ASL (25.15% for Q) and multi-delay ASL models (12.62% for perfusion). This suggests QTMnet’s accuracy in estimating perfusion parameters, indicating its potential as a clinical ASL MRI image processing tool for accurate perfusion assessment. The study emphasizes the effectiveness of deep learning-based approaches for precise perfusion estimation in medical imaging.
Original article: https://arxiv.org/abs/2311.10640