Generating synthetic contrast enhancement from non-contrast chest computed tomography using a generative adversarial...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology
• Deep Learning Applications

Background:

• Contrast-enhanced computed tomography (CECT) is crucial for evaluating mediastinal lymph nodes.
• Non-contrast CT (NCCT) is often used but offers limited visualization of these structures.
• Developing methods to enhance NCCT utility is essential for improving diagnostic workflows.

Purpose of the Study:

• To evaluate a deep learning (DL) model for synthesizing CECT (sCECT) from NCCT.
• To assess the technical image quality and clinical utility of DL-generated sCECT.
• To determine if sCECT improves the conspicuity of mediastinal lymph nodes compared to NCCT alone.

Main Methods:

• A DL model was trained and validated on three datasets for sCECT generation from NCCT.
• Technical evaluation involved image similarity metrics (MAE, PSNR, SSIM, LPIPS) on test set 1.
• Clinical utility was assessed by measuring lymph node contrast-to-noise ratio (CNR) and conducting an observer study on test set 2.

Main Results:

• sCECT demonstrated superior image quality metrics compared to NCCT, including lower MAE and higher PSNR and SSIM.
• Mediastinal lymph nodes showed significantly higher CNR in sCECT compared to NCCT (6.15 vs 0.74, P < .001).
• An observer study confirmed significantly higher lesion conspicuity with sCECT compared to NCCT alone (P ≤ .001).

Conclusions:

• Deep learning-based sCECT effectively synthesizes contrast enhancement from NCCT.
• The generated sCECT significantly improves the visualization and conspicuity of mediastinal lymph nodes.
• This AI approach holds promise for enhancing diagnostic accuracy in chest CT interpretation without additional contrast administration.