Automated Frame-by-Frame Assessment of Lung Ultrasound Imaging in Severe COVID-19 Patients Using Machine Learning

冠状病毒病（COVID-19）已成为加拿大医疗保健系统最紧迫的问题，因为每天新增病例和死亡人数都在惊人地增加。医疗机构的报告显示，在所有确诊的COVID-19患者中，约有7-14%在加拿大住院。COVID-19肺部受累是一个不祥的征兆，因为患者可以迅速失代偿，并在几个小时内需要紧急进入重症监护室（ICU）进行机械通气，以维持肺功能，直到他们康复。因此，鉴于病床和机械呼吸机等可用资源有限，准确诊断和监测患者的疾病对于更好地管理医院和ICU至关重要。肺部状况可以通过胸部X光片或计算机断层扫描（CT）进行评估。年轻患者的重复X光片或CT扫描也代表了已知导致癌症的高辐射剂量。 在所有住院的COVID-19患者中，约有12%的患者年龄小于40岁。此外，无辐射无创扫描对于怀孕的COVID-19患者至关重要。在这个项目中，我们提出使用超声成像和机器学习算法开发来检测和监测严重COVID-19患者的肺炎。拟议的项目将由阿尔伯塔大学和加拿大埃德蒙顿的MEDO.ai合作开展。超声成像是廉价的，非侵入性的，无电离辐射和便携式。由于体积小，超声波扫描仪相对容易消毒，这在高传染性COVID-19病毒的情况下至关重要。最初的机器学习算法开发将依赖于从296名患者获得的超声扫描。我们将把学习到的机器模型集成到一个基于网络的诊断系统中，以产生一个可以由受过有限培训的医疗工作者有效使用的工具。该系统还可用于医院设施有限的农村和偏远地区。

The Coronavirus disease (COVID-19) has become the most pressing concern for the Canadian healthcare system due to the alarmingly increasing number of new cases and deaths every day. Reports by healthcare agencies indicate that around 7-14% of all confirmed COVID-19 patients have been hospitalized in Canada. COVID-19 lung involvement is an ominous sign as patients can decompensate rapidly and within a matter of hours need urgent admission to an intensive care unit (ICU) for mechanical ventilation to maintain lung function until they recover. Therefore, it is critical to accurately diagnose and monitor the disease in patients for better management of hospitals and ICUs, given the limited available resources such as beds and mechanical ventilators. Lung conditions could be assessed using chest radiographs or computed tomography (CT). Repeated radiographs or CT scans of younger patients also represent a high radiation dose known to lead to cancer. About 12% of all hospital admissions for COVID-19 patients are younger than forty years old. In addition, radiation-free non-invasive scanning is essential for COVID-19 patients who are pregnant. In this project, we proposed to use ultrasound imaging with machine learning algorithm development to detect and monitor pneumonia in severe COVID-19 patients. The proposed project will be undertaken by the partnership between the University of Alberta and MEDO.ai in Edmonton, Canada. Ultrasound imaging is inexpensive, non-invasive, free of ionization radiation and portable. Due to their small size, ultrasound scanners are relatively easy to disinfect, which is critical in the case of the highly infectious COVID-19 virus. The initial machine learning algorithm development will rely on ultrasound scans obtained from 296 patients. We will integrate the learned machine models into a web-based diagnostic system to produce a tool that can be used effectively by a healthcare worker with limited training. The system could also be used in rural and remote areas with limited hospital facilities.