TRD3: Data Analytics and Intelligent Systems (AI-ML-DL-Visualization)

项目总结-技术研究与开发项目#3 在危重的临床护理方案(例如手术和重症监护)中存在未得到满足的需求，即缺乏 实时的术中成像和病理数据，用于可视化的智能系统，以及集成 多模式数据与其他临床数据一起进行实时决策指导。TRD3将发展深度学习(DL)， 机器学习(ML)、人工智能(AI)和可视化(VIS)工具来应对这些挑战。至 为了实现这一目标，研究小组将承担四个具体目标：在目标1中，研究小组将 通过联合优化光学硬件和后端机器学习模型来构建数据驱动的仪器 对于给定的任务。优化的iFLIM和iDOS仪器将具有更高的功能和更高的SNR 为临床使用而开发的。在目标2中，研究小组将开发有效和富有表现力的可视化 界面和人类对多模式成像数据的理解。这些工具将提供关键信息 用于关键决策和改进临床工作流程。在AIM 3中，研究团队将开发新的人工智能工具 整合异质多模式数据以预测患者结局。多模型数据集成 这一方法将克服孤立考虑每一种单一模式的局限性。最后，在目标4中， 研究小组将把这些新技术整合到临床工作流程中，为 手术指导。通过实现这些目标，研究团队将开发和验证一套先进的 使用AI/ML/DL的分析方法，用于智能仪器设计、数据/图像分析、可视化和临床 做决定。此TRD与TRDs1和2之间的强大交互和共享资源将使 成像和推理能力方面的性能提升。这些方法的组合将 为基于对个别患者结果的预测选择高度个性化的治疗铺平道路。

PROJECT SUMMARY – Technology Research and Development Project #3 There are unmet needs in critical clinical care scenarios (e.g., surgery and intensive care) namely the lack of real-time intraprocedural imaging and pathologic data, intelligent systems for visualization, and integration this multimodality data with other clinical data for real-time decision guidance. TRD3 will develop deep learning (DL), machine learning (ML), artificial intelligence (AI), and visualization (VIS) tools to address these challenges. To accomplish this objective the research team will undertake four Specific Aims: In Aim 1, the research team will build data-driven instruments by jointly optimizing the optical hardware and the back-end machine learning model for a given task. Optimized iFLIM and iDOS instruments with increased capabilities and higher SNR will be developed for clinical use. In Aim 2, the research team will develop effective and expressive visualization interfaces and human comprehension of multimodality imaging data. These tools will provide critical information for critical decision making and improve clinical workflow. In Aim 3, the research team will develop new AI tools to integrate heterogenous multimodality data to predict patient outcome. The multi-model data integration approach will overcome the limitations of each single modality being considered in isolation. Finally, in Aim 4, the research team will incorporate these new techniques into clinical workflow to provide real-time feedback for surgical guidance. By accomplishing these aims the research team will develop and validate a set of advanced analytical methods with AI/ML/DL for intelligent instrument design, data/image analysis, visualization, and clinical decision making. Strong interactions and shared resources between this TRD and TRDs1 and 2 will enable performance advancements in the imaging and inference capabilities. The combination of these approaches will pave the way for choosing highly personalized treatments based on predictions of individual patient outcome.