Exploration of Deep Learning Methods for Synthetic T2-Weighted Pelvic MRI Generation from CT Scans: A Technical...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Synthesizing T2-weighted Magnetic Resonance Imaging (MRI) from Computed Tomography (CT) scans is an underexplored challenge in abdominopelvic imaging.
• Deep learning approaches offer potential solutions for this ill-posed problem, aiming to bridge the gap where MRI is unavailable or contraindicated.

Purpose of the Study:

• To develop and compare deep learning algorithms for generating synthetic T2-weighted MRI from pelvic CT scans.
• To systematically evaluate different network architectures and training strategies for their feasibility and performance in CT-to-MRI synthesis.

Main Methods:

• A conditional Generative Adversarial Network (GAN) framework was employed.
• Three state-of-the-art models—efficient Self-Attention UNet (ESAUNet), Residual Vision Transformer (ResViT), and Cascaded Gaze—served as generators.
• A combined loss function (L1, VGG19 perceptual, adversarial) was used for optimization, with training on a multicenter cohort (n=90) and testing on an independent cohort (n=19).

Main Results:

• ESAUNet demonstrated the highest performance with PSNR of 22.21 dB, SSIM of 0.748, and MAE of 0.044, outperforming ResViT and CascadedGazeNet.
• Radiologists found no significant difference in overall pelvic image quality or rectal wall delineation between synthetic and true T2-weighted MRI (p > 0.29).
• Robust inter- and intra-observer agreement was achieved, despite slightly lower ratings for fine details in synthetic images.

Conclusions:

• Conditional GAN architectures are technically feasible for translating pelvic CT to T2-weighted MRI.
• ESAUNet shows significant potential for enabling MRI-equivalent imaging in resource-limited environments due to its efficiency and strong performance.
• The study validates the use of deep learning for generating synthetic MRI from CT, offering a promising avenue for enhanced diagnostic capabilities.