Enhanced Visualization: Transforming Non-Contrast into Contrast-Enhanced Computed Tomography Images Through Advanced...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Contrast-enhanced CT (CECT) is crucial for evaluating mediastinal and lymph node abnormalities.
• CECT is contraindicated in patients with renal dysfunction, contrast allergies, or in pediatric populations.
• Deep learning, specifically generative adversarial networks (GANs), offers a potential solution for generating synthetic CECT (sCECT) from non-contrast CT (NCCT).

Purpose of the Study:

• To develop and evaluate a GAN-based model for generating sCECT from NCCT.
• To assess the quantitative and qualitative performance of sCECT compared to CECT.
• To determine the utility of sCECT as an alternative in contrast- contraindications.

Main Methods:

• A GAN model was trained on 400 CECT scans.
• The model was tested on NCCT scans from 20 patients with lymphoma or metastatic lymphadenopathy.
• Quantitative metrics (MAE, RMSE, PSNR, SSIM, PCC) and qualitative assessments by radiologists were performed. Signal-to-Noise Ratio (SNR) and Contrast-to-Noise Ratio (CNR) were measured.

Main Results:

• sCECT showed modest quantitative improvements in pixel-wise similarity (lower MAE/RMSE, higher PSNR) but decreased PCC compared to CECT.
• Qualitative assessment by radiologists revealed significantly improved visualization of mediastinal structures with sCECT.
• SNR and CNR analyses indicated enhanced contrast depiction in sCECT compared to NCCT.

Conclusions:

• The GAN-based model successfully generated sCECT from NCCT, demonstrating clear qualitative improvements for mediastinal lymph node evaluation.
• While quantitative similarity gains were modest, sCECT offers a valuable adjunct for patients with contraindications to iodinated contrast agents.
• Synthetic enhancement represents a learned transformation, not true attenuation, but holds promise for specific clinical scenarios.