Development of AI/ML-ready shared repository for parametric multiphysics modeling datasets: standardization for predictive modeling of selective brain cooling after traumatic injury

开发用于参数多物理场建模数据集的 AI/ML 就绪共享存储库：创伤后选择性脑冷却预测模型的标准化

• 批准号: 10842926
• 负责人: Paolo Francesco Maccarini
• 金额: $ 30.34万
• 依托单位: DUKE UNIVERSITY
• 依托单位国家: 美国
• 财政年份: 2022
• 资助国家: 美国
• 起止时间: 2022-09-01 至 2024-08-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/10842926
• 关键词: 3-Dimensional; Acceleration; Address; Affect; Anatomy; Artificial Intelligence; Basic Science; Behavior; Big Data; Biomedical Research; Brain; Brain Injuries; Cancerous; Cardiac Surgery procedures; Catheters; Cerebrum; Clinical; Clinical Research; Code; Collaborations; Communities; Complex; Computer software; Data; Data Engineering; Data Files; Data Science; Data Set; Databases; Development; Development Plans; Device Designs; Devices; Documentation; Early Diagnosis; Electromagnetics; Engineering; Exploratory/Developmental Grant for Diagnostic Cancer Imaging; Funding; Future; Goals; Grant; Head; Health Care Costs; Information Technology; Injury; Intracranial Pressure; Learning Skill; Lesion; Location; Long-Term Effects; Machine Learning; Malignant Neoplasms; Maps; Medical Device; Medical Device Designs; Medicine; Modeling; Monitor; Outcome; Output; Patient-Focused Outcomes; Patients; Perfusion; Phase; Physics; Physiological; Power Sources; Prediction of Response to Therapy; Probability; Procedures; Process; Property; Public Health; Pythons; Quality of life; Readability; Research; Research Personnel; Research Support; Resource-limited setting; Running; Standardization; Students; System; TBI Patients; Technology; Temperature; Testing; Thermal Ablation Therapy; Time; Tissues; Training; Trauma; Traumatic injury; Treatment Protocols; United States National Institutes of Health; Validation; Variant; Ventricular; Work; aggressive breast cancer; behavior prediction; blood perfusion; brain tissue; clinical application; clinically relevant; complex data; data curation; data management; data modeling; data standards; deep learning model; design; experience; file format; graduate student; improved; innovation; insight; large datasets; learning community; machine learning algorithm; manufacture; microwave electromagnetic radiation; multidisciplinary; natural hypothermia; novel; open source; pre-clinical; predictive modeling; programs; real time monitoring; repository; response; sensor; shared repository; signal processing; simulation; skill acquisition; skills; student participation; therapy design; tool; treatment optimization; treatment planning; tumor; undergraduate student; usability

ABSTRACT By rapidly and selectively cooling injured brain tissue, we can dramatically mitigate the long-term effect of trauma to the head. As part of the NIH-funded R21, we are developing a stylet that could be easily inserted in commonly used extra ventricular catheters to add cooling to intracranial pressure control. As we are developing the device, we also realize the need for using AI/ML algorithm for optimizing design of the device and treatment planning. Unfortunately all the commercially available software that run multiphisic numerical simulation produce data that is not ready for processing by artificial intelligence and machine learning (AI/ML) technologies. Although AI/ML are data-driven technologies could potentially revolutionize biomedical research, most research data is not readily useable by AI/ML applications. In particular, there is the widespread and urgent need to make AI-ML ready the large parametric datasets generated by multiphysics numerical simulations. This supplemental project aim to address that issue and create a framework template for other clinical/basic research groups to make AI/ML ready data from complex predictive multiphysics modeling to enhance significantly their optimization and prediction capabilities. These simulations can rapidly and accurately predict the behavior of complex biomedical devices in phantom, preclinical and clinical settings. Parametric predictive multiphysics modeling (PPMM) allows researchers/clinicians/patients to study the effects of potential variations in manufacturing, treatment parameters, anatomical features and physiological responses on treatment procedures. These sensitivity studies produce significantly large datasets that could be rapidly process by AI/ML algorithms to optimize clinical procedures. As part of a recently awarded R21 grant, we are developing a new device that can rapidly and selectively cool the cerebral tissue of traumatic brain injury patients. Rapid selective brain cooling could dramatically improve patient outcomes by minimizing secondary injuries. PPMM using commercially-available software (Comsol, Ansys, Matlab, CST and others) is used both at the design stage and during the treatment planning phase. However, the significant amount of PPMM data is not ready for AI/ML processing since each 4D database lack of reference to the original set of parameter (i.e. tissue properties, perfusion rate, type and location of injury…). We thus plan, within the proposed supplemental research, to address these specific aims: 1) Develop and disseminate an AI/ML-Ready PPMM dataset 2) Demonstrate the Usability of the AI/ML-Ready PPMM dataset in an AI/ML application (optimization of treatment planning) 3) Demonstrate the usability of the AI/ML-ready PPMM dataset with student engagement activities. Although the research will be focused on brain cooling PPMM, the approach will be easily expandable to other PPMM such as cancer thermal ablation, brain temperature monitoring of hypothermic cardiac surgeries and early detection of aggressive breast cancer. The proposed research will pave the way to the full potential of AI/ML technologies in tandem with multiphysics simulations for the benefit of traumatic brain injury patients.

摘要 通过快速和选择性地冷却受伤的脑组织，我们可以大大减轻创伤的长期影响 打在头上作为NIH资助的R21的一部分，我们正在开发一种可以很容易地插入通常的探针， 使用额外的脑室导管来增加颅内压控制的冷却。在我们开发设备的过程中， 我们还认识到需要使用AI/ML算法来优化设备和治疗计划的设计。 不幸的是，所有运行多相数值模拟的商业软件产生的数据 尚未准备好通过人工智能和机器学习（AI/ML）技术进行处理。AI/ML 数据驱动的技术可能会彻底改变生物医学研究，但大多数研究数据并不 易于AI/ML应用程序使用。特别是，有广泛和迫切的需要，使AI-ML 准备好由多物理场数值模拟生成的大型参数化数据集。 这个补充项目旨在解决这个问题，并为其他临床/基础 研究小组从复杂的预测多物理场建模中获取AI/ML就绪数据，以增强 尤其是它们的优化和预测能力。这些模拟可以快速准确地预测 复杂生物医学器械在体模、临床前和临床环境中的行为。参数预测 多物理场建模（PPMM）允许研究人员/临床医生/患者研究潜在变化的影响 在制造过程中，治疗参数、解剖特征和治疗的生理反应 程序.这些敏感性研究产生了可以由AI/ML快速处理的非常大的数据集 优化临床程序的算法。作为最近授予的R21赠款的一部分，我们正在开发一种新的 一种能够快速选择性冷却创伤性脑损伤患者脑组织的装置。快速选择 大脑冷却可以通过最小化二次伤害来显著改善患者的治疗效果。 PPMM使用商业上可获得的软件（Comsol、Ansys、Matlab、CST和其他），用于 设计阶段和治疗计划阶段。然而，大量的PPMM数据并不 准备好进行AI/ML处理，因为每个4D数据库缺乏对原始参数集（即组织）的参考 性质、灌注率、损伤的类型和位置.）。因此，我们计划在拟议的补充 研究，以解决这些具体目标：1）开发和传播AI/ML就绪PPMM数据集2） 证明AI/ML就绪PPMM数据集在AI/ML应用程序中的可用性（治疗优化 3）通过学生参与活动展示AI/ML就绪PPMM数据集的可用性。 虽然这项研究将集中在大脑冷却PPMM上，但这种方法很容易扩展到其他领域。 PPMM，如癌症热消融，低温心脏手术的脑温监测和早期 检测侵袭性乳腺癌。拟议的研究将为AI/ML的全部潜力铺平道路 技术与多物理场模拟相结合，为创伤性脑损伤患者带来好处。