文献速递神经核医学肿瘤 PET/CT放射性药物与辐射防护

人工智能增强脑核医学成像

Artificial Intelligence Augmented Cerebral Nuclear Imaging.
PubMed 文献 · Semin Nucl Med · 2025-05-28

本文探讨机器学习和深度学习在脑SPECT与PET图像处理、分析、增强及解读中的应用,涵盖自动分割、疾病分类、影像组学特征提取、深度学习重建、伪CT衰减校正和低计数去噪,并讨论联邦学习与生成式AI的潜在作用。

要点

  • 机器学习与深度学习可用于脑SPECT和PET的自动分割、疾病分类及影像组学特征提取。
  • 深度学习重建、伪CT衰减校正和低计数研究去噪有助于提升图像质量。
  • 联邦学习应对数据安全挑战,生成式AI有望解决数据局限并改善患者照护与培训。

核心结论

  • 机器学习和深度学习创新正在重构核神经影像生态系统,并重新构想未来精准医学格局。

研究设计与样本

  • 本文为综述,探讨机器学习和深度学习在脑SPECT与PET图像处理、分析、增强和解读中的当前及新兴应用。

主要发现

  • 关键进展包括自动图像分割、疾病分类和影像组学特征提取,涵盖低维一阶与二阶影像组学、高维三阶影像组学及更抽象的四阶深度影像组学。
  • 基于深度学习的重建、使用伪CT生成的衰减校正以及低计数研究的去噪在提升图像质量方面具有作用。
  • 人工智能通过放射性配体设计和临床前成像在可持续性方面具有作用,联邦学习可应对数据安全挑战以改善核脑成像的研发。
  • 生成式人工智能有潜力通过解决数据局限、图像增强、以患者为中心的照护、工作流程效率和受训者教育来变革核脑成像领域。

局限性

  • 原文未报告局限性

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

原文

Artificial intelligence (AI), particularly machine learning (ML) and deep learning (DL), has significant potential to advance the capabilities of nuclear neuroimaging. The current and emerging applications of ML and DL in the processing, analysis, enhancement and interpretation of SPECT and PET imaging are explored for brain imaging. Key developments include automated image segmentation, disease classification, and radiomic feature extraction, including lower dimensionality first and second order radiomics, higher dimensionality third order radiomics and more abstract fourth order deep radiomics. DL-based reconstruction, attenuation correction using pseudo-CT generation, and denoising of low-count studies have a role in enhancing image quality. AI has a role in sustainability through applications in radioligand design and preclinical imaging while federated learning addresses data security challenges to improve research and development in nuclear cerebral imaging. There is also potential for generative AI to transform the nuclear cerebral imaging space through solutions to data limitations, image enhancement, patient-centered care, workflow efficiencies and trainee education. Innovations in ML and DL are re-engineering the nuclear neuroimaging ecosystem and reimagining tomorrow's precision medicine landscape.

来源:Semin Nucl Med,查看原文。版权归原作者所有。

引用

  1. Currie GM, Hawk KE. Artificial Intelligence Augmented Cerebral Nuclear Imaging. Semin Nucl Med. 2025. PMID: 40441996. doi:10.1053/j.semnuclmed.2025.05.005

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。