Development of COVID-19 Imaging Tools with Artificial Intelligence

在NCI TCIA公共网站上发布了CT扫描的公共数据。人工智能深度学习模型是与多个行业合作伙伴一起制作的，以了解COVID-19的连续时间动态。.人工智能深度学习模型被构建并公开发布在合作伙伴的管道上，用于研究目的，可以自动分割COVID-19不透明并在初始护理点CT扫描上对COVID-19进行分类，这些模型建立在多国爆发训练数据的基础上。模型输出是胸部CT扫描上的COVID可能性%。NIH CC和NCI是最早收集多国数据并开发基于COVID CT的免费公共AI解决方案的机构之一，供学术和商业开发人员使用。还与MICCAI和儿童国家医疗中心一起进行了一次数据挑战，供公众使用。用于临床试验环境的统一且经验证的成像生物标志物解决方案可以加快药物发现和早期验证或响应信号的途径。NIH团队正在与商业和学术合作伙伴合作，评估COVID指标的量化工具。NIH模型可以检测COVID-19，并与H1N1流感，真菌或细菌性肺炎以及癌症，正常肺和其他实体区分开来。正在进行的工作将试图识别和标记CT病例，以便立即进行放射科医生审查，从而标记和鼓励隔离，PCR检测和接触者追踪高度怀疑和/或无症状病例。其他模型根据初始护理点的初始CT扫描或胸部X光检查来预测后期对重症监护治疗的需求。标准化CT反应定量的能力将使药物组合和治疗方法之间的关键跨平台比较成为可能，这是至关重要的，因为在支持性治疗途径中需要跨药物类别的联合治疗。还表明，症状前CT AI可以以可预测的方式跟踪疾病，并且这种疾病动态曲线在COVID-19的非人灵长类动物模型中重现。先前与校外合作伙伴的工作已经证明，联邦学习可以克服成像AI的不平衡源数据的缺点，并且特定联邦学习技术的应用可以克服差距，从而表明数据不需要共享，以便从医学成像中构建高质量的AI模型。使用联邦学习的胸部X射线AI预测模型将很快发表在一本高影响力的杂志上。背景/意义：CT图像处理和深度学习模型提供了可量化的指标，可作为肺部受累的非侵入性生物标志物。与各种临床相关元数据的相关性可能使在爆发期间使用CT AI能够识别CT生物标志物特征，以便在COVID-19临床试验中进行标准化量化。这项工作与许多校园工作交叉联系，包括临床前NIAID工作和临床验证试验，用于COVID-19的分类和表征的图像处理。正在收集和整理COVID-19的多国数据集，以建立COVID-19分类和量化的公共模型，并通过分析来自4个国家的数千次CT扫描，证实无症状病毒脱落可能与阳性CT扫描共存。目标：促进标准化工具的验证，以建立用于量化的公共深度学习模型和用于表征COVID-19临床试验的标准响应标准度量。假设：CT成像数据聚合和人工智能将为COVID-19的临床和临床前研究提供信息并加快其速度。具体目标：开发、验证和翻译工具，利用深度学习方法对COVID-19疾病进行自动化和标准化CT评估和量化，以供临床试验使用。CC/NCI团队成员还在猪身上部署了一个3D打印的微型呼吸机（现已商业化），以及一个带有在线空气过滤的一次性隔离袋装置。CT AI模型被授权给工业。

Public data posting of CT scans on public NCI TCIA websites were made. AI deep learning models were made alongside of multiple industry partners, to educate on the serial temporal dynamics of COVID-19. . AI deep learning models were built and publicly posted on a partner's pipeline for research purposes that could automatically segment COVID-19 opacities and classify COVID-19 on an initial point of care CT scan, built on multi-national outbreak training data. Model output was % likelihood COVID on chest CT scans. NIH CC and NCI were among the first to gather multi-national data and develop freeware public AI solutions based on COVID CTs for both academic and commercial developer use. A data challenge was also done for public use alongside of MICCAI and Children's National Medical Center. A uniform and validated imaging biomarker solution for use for a clinical trial setting could expedite the pathway towards drug discovery and early validation or response signals. The NIH team is working with commercial and academic partners to assess quantification tools for COVID metrics. NIH models can detect COVID-19 and differentiate from H1N1 influenza, fungal, or bacterial pneumonias as well as cancer, normal lungs, and other entities with high performance. Ongoing work will attempt to identify and flag CT cases for immediate radiologist review, thus flagging and encouraging isolation, PCR testing, and contact tracing for high suspicion and or asymptomatic cases. Other models predict the later need for critical care therapies based upon an initial CT scan or chest x-ray at the initial point of care. The ability to standardize the quantification of CT responses would enable critical cross-platform comparisons among drug combinations and therapeutic approaches, which is vital, given the necessity for combination therapies across classes of drugs in supportive therapy pathways. It has also been shown that pre-symptomatic CT AI can track disease in a predictable fashion, and that this disease dynamic curve is recapitulated in a non-human primate model of COVID-19. Prior work with extramural partners has demonstrated that federated learning can overcome shortcomings in unbalanced source data for imaging AI, and that the application of a specific federated learning technique can overcome the gap, thus showing that the data does not need to be shared in order to build quality AI models from medical imaging. The chest X-ray AI predictive model using federated learning will soon be published in a high-impact journal. BACKGROUND / SIGNIFICANCE: CT image processing and deep learning models provide quantifiable metrics to serve as a noninvasive biomarker for pulmonary involvement. Correlation with a variety of clinically relevant metadata may enable the use of CT AI during outbreaks to identify CT biomarker features for standardized quantification in clinical trials for COVID-19. This effort cross links with numerous campus efforts, including preclinical NIAID efforts and clinical validation trials for image processing for classification and characterization in COVID-19. A multi-national dataset in COVID-19 is being collected and curated to build public models for COVID-19 classification and quantification and has verified that asymptomatic viral shedding may co-exist in the presence of a positive CT scan with analysis of thousands of CT scans from 4 nations. GOALS: Facilitate validation of a standardized tool for establishment of public deep learning models for quantification and standard response criteria metrics for characterization of COVID-19 clinical trials. HYPOTHESIS: CT imaging data aggregation and artificial intelligence will inform and expedite clinical and preclinical studies of COVID-19. SPECIFIC AIMS: Develop, validate, and translate tools for automated and standardized CT assessment and quantification of COVID-19 disease with deep learning methodologies for use during clinical trials. CC/NCI team members also deployed a 3D-printed miniature ventilator in swine (now commercialized) as well as a disposable isolation bag device with in-line air filtration. CT AI models were licensed to industry.