Time conditioning for arbitrary contrast phase generation in interventional computed tomography.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Minimally invasive ablation for renal cancer offers benefits like low complication rates and faster recovery.
• Computed tomography (CT) provides excellent visualization but requires iodinated contrast agents, posing risks and increasing radiation dose.
• Developing contrast-free imaging methods is crucial for interventional radiology.

Purpose of the Study:

• To investigate the use of temporal information from non-contrast CT scans to synthesize contrast-enhanced images.
• To evaluate a novel generative adversarial network (GAN) approach for creating these synthetic images during renal cryoablation.
• To assess the feasibility of using time information for contrast enhancement in interventional CT.

Main Methods:

• Proposed a novel method conditioning generative adversarial networks (GANs) with normalized time stamps.
• Employed a HyperNetwork architecture for generating synthetic contrast-enhanced CT images.
• Reduced the receptive field of the GANs to address challenges specific to interventional CT data.

Main Results:

• Generated synthetic contrast-enhanced images of competitive quality compared to standard generative models.
• Demonstrated significantly improved image quality and segmentation performance by reducing the receptive field.
• Showcased model robustness for inference on unseen intra-procedural data, improving needle artifact visualization.

Conclusions:

• Time-conditioned GANs with HyperNetworks offer a feasible approach for synthesizing contrast-enhanced CT images.
• Reducing the receptive field enhances image quality and downstream task performance in interventional CT.
• The proposed method generalizes well, improving visualization and potentially reducing contrast agent dependency in renal cryoablation procedures.