DCE-Qnet: Deep Network Quantification of Dynamic Contrast Enhanced (DCE) MRI

Area of Science:

• Medical Imaging
• Radiology
• Artificial Intelligence in Medicine

Background:

• Dynamic Contrast-Enhanced Magnetic Resonance Imaging (DCE-MRI) offers valuable clinical insights.
• Robust pharmacokinetic modeling for DCE-MRI remains a significant challenge for widespread clinical adoption.
• Accurate quantification requires complex modeling and often multiple imaging sequences.

Purpose of the Study:

• To develop and validate a novel neural network, DCE-Qnet, for comprehensive DCE-MRI quantification.
• To assess the accuracy and reproducibility of DCE-Qnet compared to conventional methods.
• To demonstrate the clinical utility and efficiency of DCE-Qnet in cancer imaging.

Main Methods:

• A 7-layer neural network (DCE-Qnet) was trained using simulated DCE-MRI data based on the Extended Tofts model and Parker arterial input function.
• Network training incorporated B1 inhomogeneities to estimate perfusion parameters (Ktrans, vp, ve), T1 relaxation, proton density, and bolus arrival time (BAT).
• Performance was evaluated in a digital phantom against nonlinear least-squares fitting (NLSQ) and in vivo in healthy subjects and a cervical cancer patient.

Main Results:

• DCE-Qnet demonstrated superior performance over NLSQ in phantom studies.
• In vivo analysis showed inter-subject variability in healthy cervix ranging from 5-51% for different parameters.
• Reproducibility in a cervical tumor was good, with coefficient of variation (CV) between 1-47%.

Conclusions:

• DCE-Qnet enables comprehensive DCE-MRI quantification from a single acquisition.
• The method eliminates the need for separate T1 scans and bolus arrival time processing, reducing scan time by approximately 10 minutes.
• DCE-Qnet offers a more accurate and efficient approach for DCE-MRI analysis, facilitating clinical adoption.