文献速递腹部技术与安全CT

利用平扫CT与深度学习人工智能机会性检测肝细胞癌

Opportunistic Detection of Hepatocellular Carcinoma Using Noncontrast CT and Deep Learning Artificial Intelligence.
PubMed 文献 · J Am Coll Radiol · 2025-01-01

本研究开发并训练了一种基于平扫多期CT的三维卷积块注意力模块(CBAM)模型用于检测肝细胞癌,共纳入2223例患者,内部验证AUC为0.807(95%CI 0.772-0.841),与放射科医生判读的0.851(95%CI 0.820-0.882)相当;外部测试集584例AUC为0.789(95%CI 0.750-0.827)。作者认为人工智能分析平扫CT在肝细胞癌机会性筛查中可能具有作用。

要点

  • CBAM模型仅用平扫CT检测肝细胞癌,内部验证AUC 0.807,与放射科判读AUC 0.851相当。
  • 在高危患者、明确结局、不确定扫描及<2 cm小病灶中,模型AUC分别为0.769、0.815、0.769、0.773。
  • 外部测试集584例中模型AUC为0.789(95%CI 0.750-0.827)。

核心结论

  • CBAM模型在内部验证中达到了与放射科判读相当的诊断准确度,人工智能分析平扫CT在肝细胞癌机会性筛查中可能具有作用。

研究设计与样本

  • 开发并训练三维卷积块注意力模块(CBAM)模型,基于平扫多期CT扫描,肝细胞癌诊断参照美国肝病研究学会指南并通过12个月临床复合参考标准确认,共纳入2223例患者,进行内部验证、独立外部测试和敏感性分析。

诊断标准与关键数值

  • CBAM模型内部验证队列AUC为0.807(95%CI 0.772-0.841),放射科判读AUC为0.851(95%CI 0.820-0.882)。
  • 高危患者中模型AUC为0.769(95%CI 0.721-0.817)。
  • 明确肝细胞癌结局病例中模型AUC为0.815(95%CI 0.778-0.853)。
  • 不确定扫描中模型AUC为0.769(95%CI 0.704-0.834)。
  • 病灶<2 cm的扫描中模型AUC为0.773(95%CI 0.692-0.854)。

临床应用与阅片要点

  • 作者认为人工智能分析平扫CT在肝细胞癌机会性筛查中可能具有作用。

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)。

英文原文(摘要)· 点击展开

OBJECTIVE: Hepatocellular carcinoma (HCC) poses a heavy global disease burden; early diagnosis is critical to improve outcomes. Opportunistic screening-the use of imaging data acquired for other clinical indications for disease detection-as well as the role of noncontrast CT have been poorly investigated in the context of HCC. We aimed to develop an artificial intelligence algorithm for efficient and accurate HCC detection using solely noncontrast CTs.

METHODS: A 3-D convolutional block attention module (CABM) model was developed and trained on noncontrast multiphasic CT scans. HCC was diagnosed following American Association for the Study of Liver Disease guidelines and confirmed via 12-month clinical composite reference standard. CT observations were reviewed by radiologists; observations in at-risk patients were annotated via the Liver Imaging Reporting and Data System. Internal validation, independent external testing, and sensitivity analyses were performed to evaluate model performance and generalizability.

RESULTS: In all, 2,223 patients were included. The CBAM model achieved an area under the receiver operating curve (AUC) of 0.807 (95% confidence interval [CI] 0.772-0.841) on the internal validation cohort, comparable to radiological interpretation at 0.851 (95% CI 0.820-0.882). Among at-risk patients, cases with definite HCC outcomes, indeterminate scans, and scans with small lesions < 2 cm in size, the model attained AUCs of 0.769 (95% CI 0.721-0.817), 0.815 (95% CI 0.778-0.853), 0.769 (95% CI 0.704-0.834), and 0.773 (95% CI 0.692-0.854). On external testing cohort with 584 patients, the CBAM model achieved an AUC of 0.789 (95% CI 0.750-0.827).

DISCUSSION: The CBAM model achieved a diagnostic accuracy comparable to radiological interpretation during internal validation. Artificial intelligence analysis of noncontrast CTs has a potential role in HCC opportunistic screening.

来源:J Am Coll Radiol,查看原文。版权归原作者所有。

引用

  1. Peng C, Yu PLH, Lu J, et al. Opportunistic Detection of Hepatocellular Carcinoma Using Noncontrast CT and Deep Learning Artificial Intelligence. J Am Coll Radiol. 2025. PMID: 40044303. doi:10.1016/j.jacr.2024.12.011

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。