SBIR Phase I Machine Learning for Screening Acute Respiratory Distress Syndrome in General and COVID-19 Patient Populations

这项小企业创新研究（SBIR）第一阶段项目的更广泛影响/商业潜力将是提高急性呼吸窘迫综合征（ARDS）的早期和准确检测。由于最近的COVID-19爆发，以及COVID-19检测呈阳性的个体发展为严重并发症的倾向，以及美国每年收治的14万名ARDS患者，ARDS检测至关重要。2018年，美国ARDS诊断市场估计为1.54亿美元。该项目将推进一种机器学习算法，以准确预测COVID-19患者群体中的ARDS发病。这些系统将监测患者的电子健康记录，并自动为普通和COVID-19患者群体提供ARDS预测警报，从而在ARDS发病之前采取适当的干预和预防方法，以改善患者的预后。这个小企业创新研究（SBIR）第一阶段项目将使用半监督机器学习（SSL）来开发和验证ARDS预测筛选工具。目标和预期的技术结果如下：目标1将采用半监督深度学习来开发一个模型，用于在发病前48小时预测ARDS。由于SSL可以提高总体性能，因此该工具可以应用于许多临床特征不可用的情况，包括缺乏放射学数据。目标2将跨外部数据集验证和优化半监督模型。对外部数据集的验证将跨各种医院特定的测量频率、人口统计数据和护理实践评估该算法。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project will be to improve early and accurate acute respiratory distress syndrome (ARDS) detection. ARDS detection is vital due to the recent COVID-19 outbreak and the propensity for individuals testing positive for COVID-19 to develop ARDS as a serious complication, as well as the 140,000 patients per year in the United States admitted with ARDS. The ARDS diagnostic market in the US was an estimated $154 million in 2018. This project will advance a machine-learning algorithm to accurately predict ARDS onset in the COVID-19 patient population. These systems will monitor patient electronic health records and automatically provide ARDS prediction alerts for both general and COVID-19 patient populations, thereby enabling appropriate intervention and prevention methods in advance of ARDS onset to improve patient outcomes. This Small Business Innovation Research (SBIR) Phase I project will use semi-supervised machine learning (SSL) to develop and validate an ARDS prediction screening tool. The goals and anticipated technical results are as follows: Aim 1 will employ semi-supervised deep learning to develop a model for the prediction of ARDS up to 48 hours prior to onset. Because SSL will improve generalized performance, the tool can be applied in settings where many clinical features are not available, including a lack of radiographic data. Aim 2 will validate and optimize the semi-supervised model across external datasets. Validation on external datasets will evaluate the algorithm across a variety of hospital-specific measurement frequencies, demographics, and care practices.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.