SCH: INT: Data-In-Motion Prediction and Assessment of Acute Respiratory Distress Syndrome

该项目的目标是开发新的计算方法，用于合成用于健康监测和早期疾病检测的实时电子健康数据流。该团队将利用这些技术来解决监测肺部疾病患者以识别急性呼吸窘迫综合征（ARDS）的问题。ARDS是一个理想的问题，因为它经常被临床医生错过，对患者造成广泛的后果。该项目将开发机器学习中的两个新兴概念，即利用特权信息和不确定性进行学习，这两个概念都与医疗保健相关。它还将开发整合不同数据类型的新方法，包括波形（例如心电图），图像（例如胸部X光）和数字数据（例如实验室结果），以更有效地帮助临床医生进行医疗诊断。该项目还将为计算机辅助健康决策支持系统建立一个多学科学习平台，为学生，博士后和早期职业临床科学家使用高效的数学工具进行精准医学做好准备。它还将包括通过招募新生，在高度多样化的实验室中整合研究培训，以及探索多学科，应用于生物医学科学和工程中现实世界问题的研究。该项目提出将机器学习技术扩展到a）将特权信息纳入算法训练（数据在回顾性数据库中常规可用，但在现场临床环境中不可用）和B）考虑训练标签中的不确定性（因为即使是医学专家在医学诊断中也具有不确定性）。这些方法将导致更准确和有效的算法，用于检测诊断不确定性常见的医疗状况。该项目将开发有效的信号处理技术，以识别时间序列数据中与呼吸功能不全和ARDS发展相关的扰动。该项目还将开发图像处理技术，从肺部的数字胸片中提取临床相关特征，从而提高许多呼吸系统疾病的实时临床诊断准确性，这些疾病通常难以区分。最后，本计画将整合这些新的方法，以发展一个临床决策支援系统。

The goal of this project is to develop new computational approaches for synthesizing streams of real-time electronic health data for health monitoring and early disease detection. The team will utilize these technologies to address the problem of monitoring patients with lung disease to identify Acute Respiratory Distress Syndrome (ARDS). ARDS is an ideal problem, because it is frequently missed by clinicians with wide-ranging consequences to patients. The project will develop two emerging concepts in machine learning, learning with privileged information and uncertainty, both of which have relevance in healthcare. It will also develop new approaches for integrating different data types, including waveforms (e.g. electrocardiograms), images (e.g. chest x-rays), and numeric data (e.g. laboratory results) to more effectively assist clinicians in medical diagnosis. The project will also establish a multidisciplinary learning platform for computer-assisted health decision support systems to prepare students, postdocs, and early career clinical scientists in precision medicine using highly effective mathematical tools. It will also include participation of groups underrepresented in STEM through recruiting new students, integrating the research training in a highly diverse laboratory, and exploring multidisciplinary, research applied to real-world problems in biomedical science and engineering.This project proposes to extend machine learning techniques to a) incorporate privileged information in algorithm training (data routinely available in retrospective databases but not live clinical environments) and b) to account of uncertainty in training labels (because even medical experts have uncertainty in medical diagnosis). These approaches will lead to more accurate and efficient algorithms for the detection of medical conditions where diagnostic uncertainty is common. The project will develop effective signal processing techniques to identify perturbations associated with respiratory insufficiency and ARDS development in time series data. The project will also develop image processing techniques that extract clinically relevant features from digital chest radiographs of the lungs that could improve the accuracy of real-time clinical diagnosis in many respiratory that are frequently difficulty to distinguish among. Finally, this project will integrate these novel methodologies to develop a clinical decision support system.