Machine learning for health modeling and simulations

用于健康建模和模拟的机器学习

• 批准号: RGPIN-2022-04462
• 负责人: deMontigny, Simon
• 金额: $ 1.82万
• 依托单位: Université de Montréal
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 财政年份: 2022
• 资助国家: 加拿大
• 起止时间: 2022-01-01 至 2023-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=760625
• 关键词: Machine; learning; health; modeling; simulations

The overarching goal of my research program is to improve and automatize the methodology of mathematical modeling in medicine and public health using artificial intelligence. I will elaborate new methods, software tools and algorithms to automatize the generation and calibration of interpretable mathematical models for decision support systems in respiratory health. The scaling of the capacity of these models, and their simulations, will enable the assimilation of data streams in real time to support stakeholders in the management of respiratory diseases and the prevention of epidemics. The three themes of my program will focus on the different aspects of automatic model generation and calibration: computer-aided composition of interpretable models (Theme 1), machine learning of model structure (Theme 2), and real-time update of simulations based on data streams (Theme 3). In Theme 1, I will use methods stemming from mathematical modeling and computer science to create flexible model templates and design a software environment to create, edit and manipulate models represented in these templates. In Theme 2, I will design algorithms to train these models to solve prediction problems represented by examples of time-dependent input-output data, and I will compare their performance to that of state-of-the-art machine learning and deep learning approaches. In Theme 3, I will design software tools for the validation and cleaning of data streams, and I will adapt active learning approaches to the process of updating simulations in real time concurrently to the data assimilation process of the underlying model. The methodological innovations of my emerging research program will result in software tools and algorithms that are urgently needed for the analysis of big health data with models that are reliable and interpretable for knowledge users as well as easy to maintain and reuse for modelers. The combination of mathematical models and machine learning in a data-driven modeling approach is a promising avenue for health research (in healthcare and public health) that has broad applicability. I have the perfect skillset to carry forward this program that draws upon the fields of mathematics, computer science and engineering, and health sciences. My research program will be supported by data obtained in interdisciplinary collaborative projects where I lead machine learning components. I will only use data that will be certified from ethics review boards, or that will be publicly available. I am already using data from the research data warehouse at Sainte-Justine's hospital pediatric intensive care unit, which has ethics certification for the development of cardiorespiratory models. I am also developing models of COVID-19 vaccination campaigns based on public health data in Quebec.

我的研究计划的总体目标是使用人工智能改进和自动化医学和公共卫生中的数学建模方法。我将阐述新的方法，软件工具和算法，以自动生成和校准呼吸健康决策支持系统的可解释的数学模型。这些模型及其模拟能力的扩大将能够在真实的时间内同化数据流，以支持利益攸关方管理呼吸道疾病和预防流行病。我的计划的三个主题将集中在自动模型生成和校准的不同方面：可解释模型的计算机辅助组成（主题1），模型结构的机器学习（主题2）和基于数据流的实时更新模拟（主题3）。在主题1中，我将使用数学建模和计算机科学的方法来创建灵活的模型模板，并设计一个软件环境来创建，编辑和操作这些模板中表示的模型。在主题2中，我将设计算法来训练这些模型，以解决由时间相关的输入输出数据示例所代表的预测问题，并将其性能与最先进的机器学习和深度学习方法进行比较。在主题3中，我将设计用于数据流的验证和清理的软件工具，并且我将使主动学习方法适应于在真实的时间中更新模拟的过程，同时适应于基础模型的数据同化过程。我的新兴研究项目的方法创新将导致软件工具和算法，这些工具和算法是分析大健康数据所迫切需要的，其模型对于知识用户来说是可靠和可解释的，并且易于维护和重用。在数据驱动的建模方法中，数学模型和机器学习的结合是健康研究（医疗保健和公共卫生）的一个有前途的途径，具有广泛的适用性。我有完美的技能来推进这一计划，借鉴了数学，计算机科学和工程领域，以及健康科学。我的研究计划将得到我领导机器学习组件的跨学科合作项目中获得的数据的支持。我将只使用将从道德审查委员会认证的数据，或将公开提供。我已经在使用来自Sainte-Justine医院儿科重症监护室的研究数据仓库的数据，该医院拥有心肺模型开发的伦理认证。我还在根据魁北克的公共卫生数据开发COVID-19疫苗接种活动的模型。