Direct Image-Based Attenuation Correction using Conditional Generative Adversarial Network for SPECT Myocardial...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Disease

Background:

• Accurate attenuation correction (AC) is crucial for quantitative analysis in SPECT myocardial perfusion imaging (MPI).
• Dedicated cardiac SPECT systems are effective for cardiovascular disease evaluation but often lack integrated CT for AC.
• Standalone SPECT scanners require alternative methods for generating attenuation maps.

Purpose of the Study:

• To develop and evaluate a conditional generative adversarial network (cGAN) for direct, one-step, CT-less attenuation correction of cardiac SPECT images.
• To generate attenuation-corrected SPECT images (SPECT_c) from non-corrected images (SPECT_nc) using a deep learning approach.
• To assess the quantitative performance of the cGAN method compared to traditional CT-based AC.

Main Methods:

• A conditional generative adversarial network (cGAN) was trained using 100 retrospective cardiac SPECT/CT datasets.
• The cGAN model generated attenuation-corrected SPECT images directly from non-corrected SPECT images.
• Quantitative evaluation involved NRMSE, PSNR, and SSIM metrics, alongside statistical analysis with joint histograms and error maps.

Main Results:

• The cGAN method achieved a 37.5% reduction in NRMSE compared to non-corrected images.
• Peak signal-to-noise ratio (PSNR) improved by 14.5%, indicating enhanced signal quality.
• Structural similarity index (SSIM) showed a 0.7% improvement, demonstrating high fidelity to reference images.

Conclusions:

• Conditional adversarial training enables accurate CT-less attenuation correction for SPECT MPI.
• The proposed method's quantitative performance is comparable to conventional CT-based attenuation correction.
• Standalone dedicated cardiac SPECT scanners can effectively utilize this GAN-based approach to mitigate attenuation artifacts.