文献速递主动脉疾病

使用人工智能对腹主动脉瘤破裂进行个体化风险评估

Individual risk assessment for rupture of abdominal aortic aneurysm using artificial intelligence.
PubMed 文献 · J Vasc Surg · 2024-12-13

本研究基于丹麦两家医院2009年1月至2016年12月的637例个体(84.8%为男性,平均年龄73±7岁,213例破裂腹主动脉瘤),采用病例对照设计开发并内部验证了SHAPFire人工智能工具,其识别破裂腹主动脉瘤的曲线下面积为0.86±0.04,显著高于仅基于直径模型的0.74±0.03(P=0.008)。结论认为该工具在识别破裂风险增高的腹主动脉瘤方面准确率显著高于单纯直径模型,但临床实施前需进行外部验证。

要点

  • SHAPFire人工智能工具纳入68个可用特征中的20个, aneurysm size、血压及身高体重关系排名最高。
  • SHAPFire模型识别破裂腹主动脉瘤的AUC为0.86±0.04,显著高于仅基于直径模型的0.74±0.03(P=0.008)。
  • SHAPFire模型性能与使用全部特征的模型相当,但临床实施前需外部验证。

核心结论

  • 本研究成功开发了SHAPFire人工智能工具,可识别破裂风险增高的腹主动脉瘤,准确率显著高于单纯直径模型。
  • 模型临床实施前需进行外部验证。

研究设计与样本

  • 采用病例对照设计的开发与验证研究,使用人工智能进行个体预后评估。
  • 纳入丹麦两家医院2009年1月至2016年12月所有可获得的破裂腹主动脉瘤病例,按1:2比例与择期手术对照匹配。
  • 排除既往腹主动脉瘤手术或缺失术前扫描的病例。
  • 样本随机均匀分为开发组和内部验证组。
  • 使用梯度提升决策树框架开发SHAPFire人工智能工具。

主要发现

  • 研究共纳入637例个体(84.8%为男性,平均年龄73±7岁,213例破裂腹主动脉瘤)。
  • SHAPFire人工智能纳入68个可用特征中的20个, aneurysm size、血压及身高体重关系获得最高排名。
  • SHAPFire模型识别破裂腹主动脉瘤的AUC为0.86±0.04,仅基于直径模型为0.74±0.03(P=0.008)。
  • SHAPFire模型性能与使用全部特征的模型相当。

临床怎么用

  • 原文未据此给出推荐。

局限性

  • 模型临床实施前需进行外部验证。

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

OBJECTIVE: This study aimed to develop a prediction tool to identify abdominal aortic aneurysms (AAAs) at increased risk of rupture incorporating demographic, clinical, imaging, and medication data using artificial intelligence (AI). DESIGN: A development and validation study for individual prognosis using AI in a case-control design. METHODS: From two Danish hospitals, all available ruptured AAA cases between January 2009 and December 2016 were included in a ratio of 1:2 with elective surgery controls. Cases with previous AAA surgery or missing preoperative scans were excluded. Features from computed tomography angiography scans and hospital records were manually retrieved. The sample was divided randomly and evenly into developmental and internal validation groups. A SHapley Additive exPlanations Feature Importance Rank Ensembling (SHAPFire) AI tool was developed using a gradient boosting decision tree framework. The final SHAPFire AI model was compared with models using (1) solely infrarenal anterior-posterior diameter and (2) all available features. RESULTS: The study included 637 individuals (84.8% men, mean age 73 ± 7 years, 213 ruptured AAAs). The SHAPFire AI incorporated 20 of 68 available features, and aneurysm size, blood pressure, and relationships between height and weight were given highest rankings. The receiver operating characteristic curve for the SHAPFire AI model displayed a significant increase in accuracy identifying ruptured AAA cases compared with the conventional model based solely on diameter with areas under the curves of 0.86 ± 0.04 and 0.74 ± 0.03 (P = .008), respectively. SHAPFire AI was comparable in performance with the model using all features. CONCLUSIONS: This study successfully developed a SHAPFire AI tool to identify AAAs at increased risk of rupture with significantly higher accuracy than diameter alone. External validation of the model is warranted before clinical implementation.

来源:J Vasc Surg,查看原文。版权归原作者所有。

引用

  1. Skovbo JS, Andersen NS, Obel LM, et al. Individual risk assessment for rupture of abdominal aortic aneurysm using artificial intelligence. J Vasc Surg. 2024. PMID: 39577479. doi:10.1016/j.jvs.2024.11.017

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。