Author: Nan Li, Yaoying Liu, Shouping Xu, Xinlei Xu, Gaolong Zhang 👨🔬
Affiliation: School of Physics, Beihang University, National Cancer Center/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, School of physics, Beihang University, Beihang University, Department of Radiation Oncology 🌍
Purpose:
CT simulation is essential for radiation therapy preparation but has limitations in distinguishing lesions. Contrast-enhanced CT (CECT) improves lesion detection and characterization, but it involves intravenous injection of iodine-based contrast agents, increasing hepatic and renal metabolic burden, raising radiation dose, and posing risks for iodine-allergic patients. This study aims to develop a deep learning model, AGCE-GAN, to enhance the contrast of normal CT scans without these drawbacks.
Methods:
The AGCE-GAN integrates a spatial attention block in the generator and a channel attention block in the discriminator. Our experimental dataset consists of CT images from 342 patients, with 270 used for model training and 72 used for model testing.
Results:
Compared with actual CECT images, AGCE-GAN-enhanced CT images show minimal differences. In arterial phase images, the mean absolute error (MAE, HU) was 30.55±4.21, peak signal-to-noise ratio (PSNR, dB) was 31.32±2.52, structural similarity index (SSIM) was 0.97±0.03, and normalized mean square error (NMSE) was 0.04±0.02. For venous phase images, the MAE(HU) was 29.55±4.11, PSNR(dB) was 30.72±2.72, SSIM was 0.97±0.01, and NMSE was 0.05±0.01. In delayed phase images, the MAE(HU) was 30.55±4.13, PSNR(dB) was 29.32±2.42, SSIM was 0.98±0.01, and NMSE was 0.04±0.01.
Conclusion:
AGCE-GAN-enhanced CT images are highly similar to actual CECT images across all phases, proving the effectiveness of our AGCE-GAN in rapidly enhancing normal CT images without contrast agents or additional scans. The high consistency with CECT images improves the detection of overlooked lesions, enhancing lesion detection rates. In radiation therapy, this method helps radiation oncologists precisely identify target margins, achieving better delineation of patient target volumes and optimizing the overall therapy process.