Automatic detection of arterial input function for brain DCE-MRI in multi-site cohorts.
Authors
Affiliations (12)
Affiliations (12)
- Department of Radiology, Loma Linda University, Loma Linda, California, USA.
- Department of Physiology and Neuroscience, Keck School of Medicine, University of Southern California, Los Angeles, California, USA.
- Zilkha Neurogenetic Institute, Keck School of Medicine, University of Southern California, Los Angeles, California, USA.
- Laboratory of NeuroImaging, USC Stevens Neuroimaging and Informatics Institute, Keck School of Medicine, University of Southern California, Los Angeles, California, USA.
- Mallinckrodt Institute of Radiology, Washington University in St. Louis School of Medicine, St. Louis, Missouri, USA.
- The Knight Alzheimer's Disease Research Center, Washington University School of Medicine, St. Louis, Missouri, USA.
- Banner Alzheimer Institute, Phoenix, Arizona, USA.
- Department of Radiology, University of Southern California, Los Angeles, California, USA.
- Alzheimer's Disease Research Center, Keck School of Medicine, University of Southern California, Los Angeles, California, USA.
- Department of Neurology, Keck School of Medicine, University of Southern California, Los Angeles, California, USA.
- Department of Psychiatry and Behavioral Sciences, University of Southern California, Los Angeles, California, USA.
- Leonard Davis School of Gerontology, University of Southern California, Los Angeles, California, USA.
Abstract
Arterial input function (AIF) extraction is a crucial step in quantitative pharmacokinetic modeling of DCE-MRI. This work proposes a robust deep learning model that can precisely extract an AIF from DCE-MRI images. A diverse dataset of human brain DCE-MRI images from 289 participants, totaling 384 scans, from five different institutions with extracted gadolinium-based contrast agent curves from large penetrating arteries, and with most data collected for blood-brain barrier (BBB) permeability measurement, was retrospectively analyzed. A 3D UNet model was implemented and trained on manually drawn AIF regions. The testing cohort was compared using proposed AIF quality metric AIFitness and K<sup>trans</sup> values from a standard DCE pipeline. This UNet was then applied to a separate dataset of 326 participants with a total of 421 DCE-MRI images with analyzed AIF quality and K<sup>trans</sup> values. The resulting 3D UNet model achieved an average AIFitness score of 93.9 compared to 99.7 for manually selected AIFs, and white matter K<sup>trans</sup> values were 0.45/min × 10<sup>-3</sup> and 0.45/min × 10<sup>-3</sup>, respectively. The intraclass correlation between automated and manual K<sup>trans</sup> values was 0.89. The separate replication dataset yielded an AIFitness score of 97.0 and white matter K<sup>trans</sup> of 0.44/min × 10<sup>-3</sup>. Findings suggest a 3D UNet model with additional convolutional neural network kernels and a modified Huber loss function achieves superior performance for identifying AIF curves from DCE-MRI in a diverse multi-center cohort. AIFitness scores and DCE-MRI-derived metrics, such as K<sup>trans</sup> maps, showed no significant differences in gray and white matter between manually drawn and automated AIFs.