Imaging biomarkers of severe respiratory infections in premature infants Phase II

摘要 早产是世界上五岁以下儿童死亡的最大单一原因， 下呼吸道感染（LRTI）是早产儿住院和死亡的首要原因。临床预测工具 早产儿LRTI的风险和严重程度评估是非常必要的，以便早期干预， 降低该患者组的高发病率和死亡率。我们的目标是改善临床实践， 制定一个客观的框架，以预测风险和评估呼吸系统疾病的严重程度， 婴儿使用非侵入性低辐射X射线成像生物标志物和临床参数。 在该项目的第一阶段，我们的多学科团队的肺科医生，肺科医生和成像和 机器学习专家开发了一种名为“肺通气和不规则”的成像软件技术 放射分析仪（LungAIR）。我们的成就包括：1）建立一个精心策划的地面真相 早产儿胸部X光检查（CXR）中的局灶性发现; 2）开发机器学习算法， 自动定位和量化基于CXR的早产儿肺病特征（纤维化/间质混浊， 囊性变化和过度充气）; 3）创建用于临床部署的图形用户界面;以及4） 在独立的队列中评估我们的成像软件技术。我们还证明了成像 LungAIR获得的生物标志物与支气管肺发育不良（BPD）的严重程度密切相关- 早产儿最常见的呼吸系统并发症--以及补充氧气的累积暴露， 新生儿重症监护室（NICU）中的机械通气（p<0.001）。重要的是，我们的初步结果 表明，影像学和临床标志物（BPD严重程度）的组合提供了准确的预测 LRTI相关并发症的模型（AUC=74，p<0.01）。 第二阶段项目以第一阶段的调查结果和方法为基础。具体目标1： 在LungAIR平台中整合肺部疾病风险因素模型。我们的软件会摄取呼吸支持 在NICU住院期间每天收集信息，并将数据与CXR分析相结合。在第二阶段，我们将 扩展LungAIR，以便在住院期间进行纵向分析，并有可能加速 预测健康风险。我们还将把我们的结果与病人的电子健康记录结合起来， 改善临床工作流程。在具体目标3中，我们将进行一项临床研究，以前瞻性地评估 LungAIR临床平台功能。该提案包括商业模式和商业化道路 呼吸机。早期识别BPD和严重LRTI高危早产儿， 结果，减少住院时间和固有的临床成本，并降低婴儿死亡率。此外该 客观量化和跟踪肺部成像生物标志物的能力也将指导治疗和临床试验， 以改善对婴儿的纵向监测。

ABSTRACT Prematurity is the largest single cause of death in children under five in the world and lower respiratory tract infections (LRTI) are the top cause of hospitalization and mortality in premature infants. Clinical tools to predict the risk and assess the severity of LRTI in premature babies are critically needed to allow early interventions to decrease the high morbidity and mortality in this patient group. Our goal is to improve clinical practice by developing an objective framework to predict the risk and assess the severity of respiratory disease in premature babies using non-invasive low-radiation X-ray imaging biomarkers and clinical parameters. In the Phase I of this project, our multidisciplinary team of pulmonologists, neonatologists and imaging and machine learning specialists developed an imaging software technology called Lung Aeration and Irregular opacities Radiological analyzer (LungAIR). Our accomplishments include: 1) establishing a curated ground truth of focal findings in chest X-Ray (CXR) of premature babies; 2) developing a machine learning algorithm to automatically localize and quantify CXR-based prematurity lung disease signatures (fibrosis/interstitial opacities, cystic changes and hyperinflation); 3) creating a graphical user interface for clinical deployment; and 4) evaluating our imaging software technology in an independent cohort. We also demonstrated that the imaging biomarkers obtained by LungAIR correlate strongly with the severity of bronchopulmonary dysplasia (BPD)—the most common respiratory complication of prematurity-- and the cumulative exposure to supplemental O2 and mechanical ventilation in the neonatal intensive care unit (NICU) (p<0.001). Importantly, our preliminary results indicated that the combination of imaging and clinical markers (BPD severity) provide an accurate predictive model for LRTI-related complications in the first year of life (AUC=74, p<0.01). This Phase II project builds on the findings and methodology developed in Phase I. In Specific Aim 1, we will incorporate a model of lung disease risk factors in LungAIR platform. Our software will ingest respiratory support information daily during NICU hospitalization and integrate the data with CXR analysis. In Specific Aim 2, we will extend LungAIR to perform longitudinal analyses during hospitalization with the potential to accelerate the prediction of health risks. We will also integrate our results with the electronic health record of the patient for improve the clinical workflow. In Specific Aim 3 we will conduct a clinical study to prospectively evaluate the LungAIR clinical platform functionality. The proposal includes the business model and a path to commercializing LungAIR. The early identification of premature babies at high risk for BPD and severe LRTI should improve their outcome, reduce hospitalization times and inherent clinical costs, and decrease infant mortality. In addition, the ability to objectively quantify and track lung imaging biomarkers will also guide therapy and clinical trials, as well as improve the longitudinal monitoring of infants.