Purpose: This study aimed to evaluate the feasibility of noise control and quantitative restoration in low-dose Positron Emission Tomography/Computed Tomography (PET/CT) images using a Generative Adversarial Network (GAN)-based Pix2Pix model, thereby assessing its clinical applicability for radiation dose reduction. Materials and Methods: Image data were acquired using a SiPM-based PET/CT scanner (GE Discovery MI) and a NEMA IEC Body Phantom. Based on the reference activity (18.50 MBq), list-mode data were collected at five reduced dose levels (5%, 10%, 25%, 50%, and 75%) according to radiopharmaceutical decay time. A total of 10,000 paired low-dose and Ground Truth (GT) dataset pairs were constructed using bootstrapping and data augmentation. Three reconstruction algorithms were evaluated: VPHD-S, VPFX-S, and QCFX-S-400. Structural and quantitative metrics, including SSIM, PSNR, Recovery Coefficient (RC), CNR, SNR, and Background Variability (BV%), were evaluated. Statistical equivalence with GT was verified using Bland-Altman analysis, Intraclass Correlation Coefficient (ICC), and paired t-tests. Statistical significance was verified using paired t-tests .Results: The Pix2Pix model significantly enhanced image quality and quantitative accuracy across all low-dose conditions. SSIM and PSNR steadily increased with dose levels, with the BSREM-based QCFX-S-400 algorithm demonstrating the highest restoration performance (SSIM: 0.967, PSNR: 37.289 dB at 75% dose). RC analysis revealed that the VPFX-S and QCFX-S-400 algorithms fully satisfied the EANM EARL2 guideline standards across all dose levels (5%–75%). Furthermore, predicted CNR and SNR were significantly higher than those of GT (). At 50% and 75% dose levels, the BV% of predicted images was lower than that of GT, demonstrating superior noise suppression. Conculsion: The Pix2Pix model successfully restored high-quality synthetic PET images while maintaining quantitative reliability even under ultra-low-dose conditions (5%–75%). Combining the deep learning model with a high-resolution reconstruction algorithm (QCFX-S-400) maximized noise reduction and quantitative accuracy for small lesions. This approach substantially reduces patient radiation dose in accordance with ALARA principles while preserving the diagnostic efficacy and safety of low-dose PET/CT image.