RAPID: Explainable Machine Learning for Analysis of COVID-19 Chest CT

2019年12月，一场广泛传染性的肺炎被发现是由一种新的冠状病毒感染引起的，现在命名为新冠肺炎。检测病毒的主要检测方法是实时聚合酶链式反应(RT-PCR)，在一些研究中灵敏度约为71%。然而，这项测试可能需要几天时间才能得出结果。也许更重要的是，需要用X光或计算机断层扫描(CT)进行成像来确认肺炎，肺炎是导致死亡的主要原因，因为它会导致急性呼吸窘迫综合征(ARDS)。最近的研究表明，胸部CT对新冠肺炎肺炎的敏感性约为98%，可以立即得出结果，但目前需要人工解读。考虑到对快速、更准确诊断的需要，本项目将使用、调整和评估可解释的机器学习技术来诊断新冠肺炎肺炎。该项目将提高对新冠肺炎发病机制的理解，并将有助于缓解其影响。使用实时聚合酶链式反应(RT-PCR)的病毒核酸检测是诊断新冠肺炎感染的主要方法，支付宝感染已作为一种全球大流行在全球迅速传播。在一些研究中，这种检测新冠肺炎感染的敏感性估计约为71%，可能需要几天时间才能得出结果。X射线和CT成像是诊断新冠肺炎肺炎的互补技术，这种肺炎可能演变为急性呼吸窘迫综合征，这是新冠肺炎感染者的主要死亡原因。尤其是在疾病的早期，胸部CT与RT-PCR相比具有多种优势，结果产生得更快，而且已经得到广泛应用，但需要放射科专家的解释。胸部CT的数量可能会迅速超过已经紧张的放射科医生的速度和容量。一种可解释的机器学习算法可以解决这一缺点，以加快胸部CT的解释，并帮助患者快速分流到ICU、住院病房、监护单元或家庭自我隔离。机器学习算法，特别是那些利用深度卷积神经网络(深度学习)的算法，有可能在几分钟内实现更快的诊断。该项目旨在验证可解释的深度学习方法的使用，以调整多种应用的诊断操作点，包括(A)疾病筛查、(B)疾病分期和预测以及(C)治疗反应评估。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

In December 2019, it was discovered that a widely contagious pneumonia was caused by a new coronavirus infection now named COVID-19. The primary test for detection of the virus is real-time polymerase chain reaction (RT-PCR) with sensitivity of approximately 71% in some studies. However, this test may require several days to provide a result. Perhaps more importantly, imaging with x-ray or computed tomography (CT) are required to confirm pneumonia, which is the principal cause of death, as it leads to acute respiratory distress syndrome (ARDS). Recent studies have shown sensitivity of chest CT for approximately 98% for COVID-19 pneumonia and could provide immediate results but currently require human interpretation. Given the need for rapid, more accurate diagnosis, this project will use, adapt, and evaluate explainable machine learning techniques to diagnosis of COVID-19 pneumonia. This project will improve the understanding of mechanisms of COVID-19 and will help mitigate its impacts.Viral nucleic acid detection using real-time polymerase chain reaction (RT-PCR) is the primary method for diagnosis of COVID-19 infection, which has rapidly spread worldwide as a global pandemic. Sensitivity of this test for COVID-19 infection has been estimated at approximately 71% in some studies and may require several days for a result. X-ray and CT imaging are complementary technologies that allow diagnosis of COVID-19 pneumonia, which can evolve to acute respiratory distress syndrome (ARDS) -- the principal cause of death in patients with COVID-19 infection. Especially early in the course of the disease, chest CT has multiple advantages over RT-PCR yielding results more quickly and is already widely deployed, but requires expert radiologist interpretation. The number of chest CTs may rapidly exceed the speed and capacity of already strained radiologists. An explainable machine learning algorithm may address this disadvantage to expedite the interpretation of chest CT and assist rapid triage of patients to the ICU, inpatient ward, monitoring unit, or home self-quarantine. Machine learning algorithms, specifically those leveraging deep convolutional neural networks (deep learning), have the potential for facilitating even more rapid diagnosis within minutes. This project seeks to validate the use of explainable deep learning methods to adjust diagnostic operating points for multiple applications, including (a) disease screening, (b) disease staging and prognostication, and (c) evaluation of treatment response.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.