基于自动化深度学习主动脉分区分割模型对急性非复杂性B型主动脉夹层进行体积分析
本研究利用自动深度学习模型,按血管外科学会(SVS)/胸外科学会(STS)定义的主动脉分区,对急性非复杂性B型主动脉夹层(auTBAD)患者进行主动脉分区体积分析;快速生长组(33例)与无或缓慢生长组(26例)基线各分区体积均无差异,模型平均Dice系数0.73,分区5和9表现最佳。
要点
- 构建了自动深度学习主动脉分区分割模型,平均Dice系数0.73,分区5(0.84)和9(0.91)表现最佳。
- 59例中快速生长33例(56%)、无或缓慢生长26例(44%),两组基线各分区体积均无差异(P > .05)。
- 全部患者中分区4(13.9%,IQR -6.82至35.1)和分区5(13.4%,IQR -7.78至37.9)体积相对增幅最大。
核心结论
- 作者描述了一个纳入SVS定义主动脉分区的自动深度学习分割模型,开源且训练后的模型与手动分割主动脉具有一致性,分区5和9表现最强,为进一步临床应用提供了框架;在有限样本中,快速生长与无或缓慢生长患者的基线主动脉分区体积无差异。
研究设计与样本
- 研究纳入具有连续影像的auTBAD患者,共59例,分析基线入院时CT血管造影(CTA)与最近一次监测CTA或主动脉干预前最近一次CTA。
- 按主动脉生长速度将患者分为快速生长组(直径增加≥5 mm/年)与无或缓慢生长组(直径增加<5 mm/年)。
- 去标识图像导入开源医学图像分析软件包,按SVS/STS标准对主动脉分区进行标注,模型采用四折交叉验证训练,分割输出用于计算各次影像的主动脉分区体积。
主要发现
- 59例中快速生长33例(56%),无或缓慢生长26例(44%)。
- 两组在基线人口学特征、合并症、入院平均动脉压、出院抗高血压药数量及高危影像特征方面均无差异(P > .05)。
- 基线至间隔CT的中位时间为1.07年(IQR 0.38-2.57)。
- 出院后13例(22%)接受主动脉干预,平均时间1.5 ± 1.2年,两组间无差异(P > .05)。
- 全部患者中分区4(13.9%;IQR -6.82至35.1)和分区5(13.4%;IQR -7.78至37.9)体积随时间相对增幅最大;两组基线各分区体积均无差异(P > .05)。
局限性
- 作者指出本研究样本有限(In our limited sample)。
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
BACKGROUND: Machine learning techniques have shown excellent performance in three-dimensional medical image analysis, but have not been applied to acute uncomplicated type B aortic dissection (auTBAD) using Society for Vascular Surgery (SVS) and Society of Thoracic Surgeons (STS)-defined aortic zones. The purpose of this study was to establish a trained, automatic machine learning aortic zone segmentation model to facilitate performance of an aortic zone volumetric comparison between patients with auTBAD based on the rate of aortic growth. METHODS: Patients with auTBAD and serial imaging were identified. For each patient, imaging characteristics from two computed tomography (CT) scans were analyzed: (1) the baseline CT angiography (CTA) at the index admission and (2) either the most recent surveillance CTA or the most recent CTA before an aortic intervention. Patients were stratified into two comparative groups based on aortic growth: rapid growth (diameter increase of ≥5 mm/year) and no or slow growth (diameter increase of <5 mm/year). Deidentified images were imported into an open source software package for medical image analysis and images were annotated based on SVS/STS criteria for aortic zones. Our model was trained using four-fold cross-validation. The segmentation output was used to calculate aortic zone volumes from each imaging study. RESULTS: Of 59 patients identified for inclusion, rapid growth was observed in 33 patients (56%) and no or slow growth was observed in 26 patients (44%). There were no differences in baseline demographics, comorbidities, admission mean arterial pressure, number of discharge antihypertensives, or high-risk imaging characteristics between groups (P > .05 for all). Median duration between baseline and interval CT was 1.07 years (interquartile range [IQR], 0.38-2.57). Postdischarge aortic intervention was performed in 13 patients (22%) at a mean of 1.5 ± 1.2 years, with no difference between the groups (P > .05). Among all patients, the largest relative percent increases in zone volumes over time were found in zone 4 (13.9%; IQR, -6.82 to 35.1) and zone 5 (13.4%; IQR, -7.78 to 37.9). There were no differences in baseline zone volumes between groups (P > .05 for all). The average Dice coefficient, a performance measure of the model output, was 0.73. Performance was best in zone 5 (0.84) and zone 9 (0.91). CONCLUSIONS: We describe an automatic deep learning segmentation model incorporating SVS-defined aortic zones. The open source, trained model demonstrates concordance to the manually segmented aortas with the strongest performance in zones 5 and 9, providing a framework for further clinical applications. In our limited sample, there were no differences in baseline aortic zone volumes between patients with rapid growth and patients with no or slow growth.
引用
- Krebs JR, Imran M, Fazzone B, et al. Volumetric analysis of acute uncomplicated type B aortic dissection using an automated deep learning aortic zone segmentation model. J Vasc Surg. 2024. PMID: 38851467. doi:10.1016/j.jvs.2024.06.001
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。