Spatiotemporal denoising of low-dose cardiac CT image sequences using RecycleGAN

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Electrocardiogram (ECG)-gated multi-phase computed tomography angiography (MP-CTA) is crucial for diagnosing coronary artery disease.
• Reducing radiation dose in MP-CTA is essential due to the need for wide cardiac phase coverage.
• Current dose reduction techniques involve acquiring only a portion of cardiac phases at full dose.

Purpose of the Study:

• To develop a spatiotemporal deep learning method for enhancing low-dose CTA images.
• To improve image quality at reduced radiation dose phases in MP-CTA.
• To introduce RecycleGAN, an advancement over CycleGAN for temporal denoising.

Main Methods:

• Developed RecycleGAN, a recurrent network-based deep learning model, to translate low-dose to full-dose image sequences.
• Utilized the XCAT phantom program for realistic MP-CTA image sequence generation for training and testing.
• Evaluated RecycleGAN's denoising performance against CycleGAN using simulated and clinical MP-CTA datasets.

Main Results:

• RecycleGAN demonstrated superior denoising performance compared to CycleGAN in simulated MP-CTA images.
• Quantitative metrics and visual inspection confirmed RecycleGAN's enhanced image quality.
• Clinical MP-CTA images further validated the superior denoising capabilities of RecycleGAN.

Conclusions:

• RecycleGAN effectively enhances the quality of low-dose CTA images by leveraging temporal information.
• The proposed spatiotemporal deep learning method offers a promising solution for radiation dose reduction in MP-CTA.
• RecycleGAN represents a significant advancement in denoising low-dose CT images for improved cardiac imaging.