Privacy-preserving Virtual Contrast-enhanced MRI for Nasopharyngeal Carcinoma: A Multicenter Study

Area of Science:

• Artificial Intelligence in Medical Imaging
• Radiology and Medical Imaging
• Oncology

Background:

• Contrast-enhanced MRI is vital for nasopharyngeal carcinoma (NPC) management.
• Gadolinium-based contrast agents pose risks, especially for patients with kidney issues.
• A safer alternative for MRI contrast enhancement is needed.

Purpose of the Study:

• To develop and assess the clinical feasibility of a federated learning model for synthesizing virtual contrast-enhanced MRI (VCE-MRI) images.
• To create VCE-MRI from contrast-free scans for NPC patients, avoiding gadolinium contrast agents.
• To provide a safer imaging alternative for NPC diagnosis, staging, and treatment.

Main Methods:

• Developed and evaluated a federated learning-based VCE-MRI synthesis model (FL-VCE-MRI).
• Trained the model on pre-treatment contrast-free T1-weighted and T2-weighted MRI scans from 14 centers (2,061 patients).
• Validated externally using data from 25 centers (126 patients) and assessed by ten clinicians.

Main Results:

• The FL-VCE-MRI model demonstrated high generalizability with a mean absolute error (MAE) of 45.85 in external validation.
• Improved average MAE from 53.93 to 45.93 for centers with challenging single-center models.
• Synthesized VCE-MRI images were deemed reliable and clinically valuable, comparable to gadolinium-enhanced MRI for diagnosis, staging, and delineation.

Conclusions:

• The FL-VCE-MRI model shows significant potential as a non-invasive alternative to gadolinium-enhanced MRI for NPC patients.
• This approach addresses the need for safer contrast enhancement in MRI for cancer imaging.
• Federated learning enables robust VCE-MRI synthesis across multiple institutions.