文献速递胸部心血管技术与安全CTDR

基于深度学习在胸部X线片上检测升主动脉扩张的诊断研究

Deep learning-based detection of ascending aortic dilatation on chest radiographs: A diagnostic study.
PubMed 文献 · Eur J Radiol · 2025-08-25

本研究采用两个回顾性外部测试集(Group 1 配对CXR/CT间隔30天内,336例;Group 2 间隔90天内,190例),评估AI算法(DeepCatch X Aorta v1.1.0)在PA位胸片上诊断升主动脉扩张(AAD)的性能,以CT半自动测量固定阈值(4.0、4.5、5.0 cm)及2名心血管放射科医师和3名低年资医师的主观评估为参照。AI在≥4 cm阈值下Group 1的AUC为0.89(95%CI 0.85,0.92),Group 2为0.88(95%CI 0.83,0.92),所有阅片者AUC均低于AI;阅片者间一致性为一般至中等(Group 1 Cohen's kappa 0.38;Group 2 Fleiss's kappa 0.37)。

要点

  • AI在PA胸片上检测升主动脉扩张≥4 cm阈值的AUC在Group 1为0.89(95%CI 0.85,0.92),Group 2为0.88(95%CI 0.83,0.92)。
  • 在4 cm阈值下,所有阅片者(含>20年经验心血管放射科医师)的AUC均低于AI。
  • 阅片者间一致性为一般至中等:Group 1 Cohen's kappa 0.38,Group 2 Fleiss's kappa 0.37。

核心结论

  • AI算法可从单张PA位胸片检测升主动脉扩张,在不同临床阈值和队列中AUC均超过0.8,表现优于人类阅片者。

研究设计与样本

  • 诊断研究,采用两个回顾性收集的外部测试集:Group 1为配对CXR/CT间隔30天内的336例扫描(平均年龄64岁),Group 2为间隔90天内的190例扫描(平均年龄70岁);以CT半自动测量固定阈值(4.0、4.5、5.0 cm)及2名>20年经验心血管放射科医师和3名<2年经验医师的主观评估为参照。

诊断标准与关键数值

  • 升主动脉扩张定义为CT测量≥4 cm,Group 1中182例、Group 2中94例存在扩张。AI在≥4 cm阈值下Group 1的AUC为0.89(95%CI 0.85,0.92),Group 2的AUC为0.88(95%CI 0.83,0.92)。在4 cm阈值下所有阅片者AUC均低于AI。Group 1中分析了校正X线放大效应的修正阈值。

临床应用与阅片要点

  • 原文未据此给出推荐。

局限性

  • 两个测试集均为回顾性收集。

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

OBJECTIVES: This study externally tests the performance of an artificial intelligence algorithm (AI) for diagnosing ascending aortic dilatation (AAD) using PA view chest radiography (PA CXR).

MATERIALS AND METHODS: Two retrospectively collected cohorts with paired CXR/CT within 30 days (Group 1) and 90 days (Group 2) were gathered as external test sets. The performance of AI (DeepCatch X Aorta v1.1.0) for detecting AAD using PA CXR was analyzed against semi-automatic measurements on CT using fixed thresholds (4.0, 4.5, and 5.0 cm) and subjective assessment (dilatation or no dilatation) of 2 cardiovascular radiologists (>20 years' experience) and 3 less-experienced (<2 years' experience) doctors. Additionally, modified thresholds accounting for radiographic magnification were analyzed in the Group 1. Performance metrics (sensitivity, specificity, and AUC) and inter-reader agreement were used.

RESULTS: Group 1 included 336 scans (mean age = 64 years) and Group 2 had 190 scans (mean age = 70 years). Dilatation (≥4 cm) was present in 182 patients (Group 1) and 94 patients (Group 2). The algorithm demonstrated robust performance in both Groups for the ≥4 cm threshold, with AUC of 0.89 (95 % CI: 0.85, 0.92) in Group 1 and 0.88 (95 % CI: 0.83, 0.92) in Group 2. All readers' performances were under the AUC of AI at 4 cm threshold. Inter-reader agreement was fair to moderate (Cohen's kappa, Group 1, 0.38; Fleiss's kappa, Group 2, 0.37).

CONCLUSION: The AI algorithm detected AAD from a single PA CXR with robust AUC over 0.8 across different clinical thresholds and cohorts, outperforming human readers.

来源:Eur J Radiol,查看原文。版权归原作者所有。

引用

  1. Lee H, Park G, Park C, et al. Deep learning-based detection of ascending aortic dilatation on chest radiographs: A diagnostic study. Eur J Radiol. 2025. PMID: 40907419. doi:10.1016/j.ejrad.2025.112380

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。