Combination of deep learning reconstruction and quantification for dynamic contrast-enhanced (DCE) MRI

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Radiology

Background:

• Dynamic Contrast-Enhanced MRI (DCE-MRI) is crucial for assessing tumor vascularity, extent, and treatment response.
• Clinical use of quantitative DCE-MRI is hindered by acquisition/quantification challenges and lack of automation.
• Current reconstruction methods, like compressed sensing, are time-consuming.

Purpose of the Study:

• To develop and validate an end-to-end deep learning pipeline for fast and quantitative DCE-MRI.
• To address limitations in speed and quantification robustness of existing DCE-MRI techniques.
• To enable improved characterization of tumor behavior and treatment efficacy.

Main Methods:

• Developed DCE-Movienet for rapid reconstruction of high spatiotemporal resolution 4D MRI data (0.66s reconstruction time).
• Integrated DCE-Movienet with DCE-Qnet for comprehensive quantification of perfusion parameters (Ktrans, vp, ve) and other factors (T1, B1, BAT).
• Processed data using a golden-angle stack-of-stars k-space trajectory and validated against compressed sensing.

Main Results:

• Achieved reconstruction times of 0.66s, a significant improvement over compressed sensing (approx. 10 min).
• Maintained image quality while drastically reducing reconstruction time.
• Demonstrated comprehensive quantification of key DCE-MRI parameters.

Conclusions:

• The end-to-end deep learning pipeline offers a substantial advancement in DCE-MRI speed and quantification.
• This technique is expected to enhance the clinical utility and performance of DCE-MRI.
• Addresses critical barriers to the widespread adoption of quantitative DCE-MRI.