Fully automated on-scanner aortic four-dimensional flow magnetic resonance imaging processing and hemodynamic...

Area of Science:

• Cardiovascular Imaging and Hemodynamics
• Medical Physics and Engineering
• Artificial Intelligence in Healthcare

Background:

• Four-dimensional (4D) flow cardiovascular magnetic resonance (CMR) is crucial for assessing cardiovascular hemodynamics.
• Traditional 4D flow CMR analysis requires time-consuming offline preprocessing and segmentation.
• This limits the immediate clinical utility of the technique.

Purpose of the Study:

• To integrate 4D flow processing tasks directly into the MRI scanner reconstruction pipeline.
• To develop an automated, on-scanner workflow for hemodynamic analysis.
• To evaluate the performance and efficiency of the automated workflow compared to manual analysis.

Main Methods:

• The Framework for Image Reconstruction (FIRE) was utilized to build containerized applications for 4D flow processing.
• Deep learning models for pre-processing, aorta segmentation, and velocity mapping were implemented in TensorFlow.
• These tasks were executed in-line on the MRI scanner using its computational resources, directly after data acquisition.

Main Results:

• On-scanner automated 4D flow hemodynamic analysis was successfully implemented on a 1.5T MRI system.
• Total on-scanner computation time was 220 ± 35 seconds.
• Deep learning models demonstrated good to excellent performance in segmentation and analysis tasks, with minimal bias in peak systolic velocity quantification compared to manual analysis.

Conclusions:

• An automated 4D flow processing workflow was successfully deployed for fully automated on-scanner hemodynamic analysis.
• The in-line analysis showed good performance comparable to human analysis.
• This automated approach has the potential to significantly increase workflow efficiency in cardiovascular imaging.