Data-Driven Approaches to Identify Biomarkers for Guiding Coronary Artery Bifurcation Lesion Interventions from Patient-Specific Hemodynamic Models

摘要 冠状动脉疾病（CAD）在美国非常流行，2017年造成超过36万人死亡 一个人CAD是由斑块（a.k.a.病变），这些病变沿着冠状动脉壁积聚，限制了 血液流动。在20%的病例中，这些病变发生在动脉分叉处。冠状动脉分叉病变的治疗 由于其支架植入具有较高的不良心脏事件风险， 支架内再狭窄、支架内血栓形成、心肌梗死或需要复发性经皮冠状动脉介入治疗， 预防（PCI）。对于单支血管病变（不在分叉处），血流储备分数与血管造影术 多支血管评价（FAME）试验在建立生物标志物（血流储备分数，FFR）方面发挥了关键作用 来指导和改善他们的治疗。然而，迫切需要一个分类方案， 分叉处病变的生理严重程度和缺血负荷，特别是主干术后侧支 分支干预。在这一知识缺口得到纠正之前，分叉病变患者将继续存在 长期心脏并发症的发生率显著高于单支主要分支病变患者。 基于FFR的当前PCI方案用于治疗较简单的主分支病变， 更复杂分叉病变的治疗方案。提取类似指标的困难是由于 病变几何形状的复杂性增加（通常由两个不同的病变组成，一个在主分支中， 一个在侧分支中），并且对下面的患者解剖结构的顺应性更强。虽然众所周知，治疗- 切除主分支病变可改善预后，但何时治疗分支病变缺乏明确的指导 分支。我们的长期目标是建立一个基于病变和患者特异性的多层次分类系统。 这些功能可用于更精确地指导治疗决策，并最终降低高风险。 分叉病变患者的不良并发症发生率。我们的中心假设是， 可以识别分叉病变解剖结构，对缺血负荷进行分类，进而指导支架植入决策。 通过使用一个系统的，经过验证的计算模型，我们现在可以准确地确定控制， 每个解剖特征与生理严重度的比例。我们现在有了计算能力、经过验证的工具， 机器学习成熟度需要进行大规模的计算机研究，不仅要隔离 单个的特征，但是特征集合之间的潜在关系。这项建议的主要目的是 通过识别病变特异性特征， 这会影响功能严重程度以及可能加剧负担的患者特异性生物标志物。

ABSTRACT Coronary artery disease (CAD) is highly prevalent in the US, causing more than 360,000 deaths in 2017 alone. CAD is caused by plaques (a.k.a. lesions) that build up along the walls of coronary arteries, restricting blood flow. In 20% of cases, these lesions occur at arterial bifurcations. Treatment of coronary bifurcation le- sions remains particularly challenging, as their stenting carries a higher risk for adverse cardiac events such as in-stent restenosis, stent thrombosis, myocardial infarction, or need for recurrent percutaneous coronary inter- vention (PCI). For single vessel lesions (not at bifurcations), the Fractional Flow Reserve Versus Angiography for Multivessel Evaluation (FAME) trial played a critical role in establishing a biomarker (fractional flow reserve, FFR) to guide and improve their treatment. However, there is an urgent need for a classification scheme to assess physiological severity and ischemic burden of lesions at bifurcations, particularly in the side branches after main branch intervention. Until this knowledge gap is corrected, patients with bifurcation lesions will continue to have a significantly higher rate of long-term cardiac complications compared to those with single, main branch lesions. Current PCI protocols based on FFR for treating simpler main branch lesions do not translate into effective protocols for more complicated bifurcation lesions. The difficulty in extracting similar metrics is due to the in- creased complexity of the lesion geometry (typically consisting of two distinct lesions, one in the main branch and one in the side branch) and stronger influence of the underlying patient anatomy. While it is known that treat- ing the main branch lesion can improve the outcome, clear guidance is lacking regarding when to treat the side branch. Our long-term goal is to establish a multi-level classification system based on lesion- and patient-specific features that can be used to guide treatment decisions with better precision, and ultimately to reduce the high rate of adverse complications in patients with bifurcation lesions. Our central hypothesis is that criteria describing bifurcation lesion anatomy can be identified to classify ischemic burden and, in turn, guide stenting decisions. Through the use of a systematic, validated computational model, we can now accurately determine the contri- bution of each anatomic feature to physiologic severity. We now have the computing power, validated tools, and machine learning maturity required to undertake a large-scale, in silico study to isolate not only the influence of individual features, but underlying relationships between sets of features. The major objective of this proposal is to enable personalized guidance of bifurcation stenting procedures by identifying both the lesion-specific features that influence functional severity as well as the patient-specific biomarkers that may exacerbate burden.