Low-dose static grating-based X-ray tri-contrast computed tomography with a contrast ranking-driven network

Grating-based X-ray computed tomography (GBCT) can obtain absorption, dark-field, and phase contrast, which offer high visibility for high-density materials, soft tissues, and microstructures, respectively. However, even with the implementation of techniques like static imaging to lower the dosage, the dosage in GBCT is still too high for clinical application. This study ranks tri-contrast X-ray CT images, incorporating absorption, dark-field, and phase contrasts, based on their performance in low-dose imaging. Subsequently, a Tri-contrast Ranking-driven Conditional Generative Adversarial Network (TR-CGAN) is presented to improve the quality of low-dose tri-contrast X-ray CT images. In TR-CGAN, the Feature Maps Prior Learning Module facilitates the enhancement of the target contrast by utilizing complementary information from other higher-quality contrasts. Furthermore, a loss function with gradient and perception is constructed for TR-CGAN enhancement. The TR-CGAN is validated with low-dose experiments. In an experiment utilizing a combination of low tube current, sparse-view, and static imaging techniques, the TR-CGAN improved the tri-contrast Feature Similarity Index Measure by 12.26%, 7.04%, and 22.18%, respectively, and the Peak Signal-to-Noise Ratio by 19.2338 dB, 3.7054 dB, and 8.2852 dB. These improvements collectively demonstrate that the proposed method enables grating-based X-ray tri-contrast imaging to yield high-quality images at substantially reduced radiation dose and acquisition time, thereby fulfilling stringent clinical requirements and furnishing more comprehensive, high-fidelity images for efficient and accurate diagnosis.