Clinical Decision Support for Assessing Pulmonary Embolism using Machine Learning

项目概要/摘要 Minnesota HealthSolutions (MHS) 提出了一个第二阶段项目来开发和验证软件产品 能够使用临床常规肺栓塞自动检测和分期肺栓塞 (PE) CT 血管造影 (CTA)。所提出的系统将结合最先进的机器学习方法和 将杜克大学的临床专业知识融入到一个无缝集成到放射学工作流程中的系统中， 标准的患者护理路径，以改善急诊科医生的治疗决策。 肺栓塞是医院患者死亡的第三大常见原因，估计 发病率为每 1,000 名患者 1 人。如今，CTA 通常用于检测 PE；然而，有显着的 放射科医生使用 CTA 的检测率存在差异。此外，尽管有强有力的证据表明 RV/LV 比值是一个重要的临床生物标志物，在实践中很少进行定量测量。一个成功的 该项目的完成将提供一个工作流程集成工具，能够实现更快的 PE 检测和更多功能 右心紧张的准确分期可指导医生的治疗决策。

Project Summary/Abstract Minnesota HealthSolutions (MHS) proposes a Phase II project to develop and validate a software product capable of automatically detecting and staging pulmonary embolisms (PEs) using clinically routine pulmonary CT angiograms (CTAs). The proposed system will combine state-of-the-art machine learning methods and the clinical expertise at Duke University into a system that integrates seamlessly into the Radiology workflow and standard patient care path to improve the treatment decisions of physicians in the emergency department. Pulmonary embolism is the third most common cause of death in hospital patients with an estimated incidence of 1 per 1,000 patients. CTAs are routinely used to detect PE today; however, there is significant variability in the detection rate among radiologists using CTA. Furthermore, despite the strong evidence that the RV/LV ratio is an important clinical biomarker it is rarely measured quantitatively in practice. A successful completion of this project would provide a workflow-integrated tool capable of faster PE detection and more accurate staging of right heart strain to guide the physician’s treatment decision.