Virtual Multi-Phase Contrast Enhanced Liver MRI Using Deep Learning for Evaluating Hepatocellular Carcinoma

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Hepatocellular carcinoma (HCC) diagnosis relies on contrast-enhanced MRI (CE-MRI).
• Traditional multi-phase CE-MRI requires significant time and Gadolinium-based contrast agents (GBCAs).
• Deep learning (DL) offers potential for improving MRI acquisition efficiency and safety.

Purpose of the Study:

• To develop and evaluate a DL model for synthesizing multi-phase CE-MRI.
• To assess the image quality, diagnostic performance, and LI-RADS utility of DL-synthesized CE-MRI for HCC detection.
• To compare the efficiency and safety of DL-based CE-MRI with conventional methods.

Main Methods:

• A DL model was trained on CE-MRI data from 717 patients with HCC or other liver diseases.
• The model synthesized arterial, portal venous, transitional, and hepatobiliary phase CE-MRI.
• Three radiologists evaluated image quality, diagnostic performance, LI-RADS features, and artifacts of DL-synthesized versus actual CE-MRI.

Main Results:

• DL-synthesized CE-MRI showed non-inferior image quality and diagnostic performance for HCC detection compared to actual CE-MRI.
• Excellent agreement was observed for LI-RADS major features between DL-synthesized and actual CE-MRI.
• The DL model generated multi-phase CE-MRI in seconds, drastically reducing acquisition time (0.20–0.60s vs. >20 min) and eliminating GBCA use.

Conclusions:

• The DL model effectively synthesizes multi-phase CE-MRI with high image quality and diagnostic accuracy for HCC.
• This AI-driven approach offers significant time savings, eliminates the need for contrast agents, and demonstrates robustness.
• The DL-based strategy shows strong potential for clinical translation, improving patient care and healthcare efficiency in liver imaging.