Segmentation Guided Crossing Dual Decoding Generative Adversarial Network for Synthesizing Contrast-Enhanced Computed...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Contrast-enhanced computed tomography (CE-CT) is crucial for diagnosing focal liver lesions (FLLs).
• Contrast agents pose a physical burden on patients.
• Generative models can synthesize CE-CT images from non-contrast CT, but often neglect critical regions.

Purpose of the Study:

• To develop an innovative CE-CT image synthesis model, SGCDD-GAN, to overcome limitations of existing methods.
• To improve the accuracy of downstream tasks like FLL classification using synthesized CE-CT images.

Main Methods:

• Proposed the Segmentation Guided Crossing Dual Decoding Generative Adversarial Network (SGCDD-GAN).
• Utilized a crossing dual decoding generator with an attention decoder and an improved transformation decoder.
• Employed a multi-task learning strategy to focus on lesion areas.

Main Results:

• SGCDD-GAN demonstrated superior performance in synthesizing both arterial (ART) and portal venous (PV) phase CE-CT images.
• Achieved high scores for SSIM, PSNR, MSE, and PCC across the entire image and liver region.
• Synthesized images led to high accuracy rates (82.68%-94.11%) in a deep learning-based FLL classification task.

Conclusions:

• SGCDD-GAN effectively synthesizes high-quality CE-CT images, highlighting critical regions.
• The model shows significant potential for improving FLL diagnosis and reducing patient burden.
• The synthesized images are valuable for deep learning-based FLL classification tasks.