摘要胸部低剂量计算机断层扫描（LDCT）因辐射剂量低而广泛用于肺癌高危人群筛查及随访，但图像噪声增加和对比度下降等问题限制了其诊断效能。深度学习技术通过数据驱动的方式，在胸部LDCT图像去噪中展现出突破性潜力。主要介绍了监督学习、无监督学习和自监督学习等深度学习模型的优势与局限，并分析其在临床应用中的潜力与挑战，以期为后续研究和临床实践提供参考。

abstractsChest low-dose computed tomography (LDCT) is a widely utilized modality for lung cancer screening and follow-up in high-risk populations, owing to its low radiation dose. However, the diagnostic accuracy of LDCT is significantly constrained by inherent limitations, including elevated image noise and reduced contrast resolution. The potential for deep learning technologies to address these challenges through data-driven LDCT image denoising approaches has been demonstrated. In this review, the advantages and limitations of deep learning models were introduced, including supervised, unsupervised, and self-supervised learning. The potential and challenges of these models in clinical applications were analyzed, thereby providing a reference for subsequent research and clinical practice.