TOPIC 427 - De-Identification Software Tools and Pipelines for Cancer Imaging Research

主题 427 - 用于癌症成像研究的去识别化软件工具和管道

• 批准号: 10612710
• 负责人: LAWRENCE O'SULLIVAN
• 金额: $ 4.36万
• 依托单位: IMPACT BUSINESS INFORMATION SOLUTIONS INC.
• 依托单位国家: 美国
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-09-16 至 2022-09-15
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/10612710
• 关键词: Address; Algorithms; Clinical; Complex; Computer software; Data; Data Element; Development; Digital Imaging and Communications in Medicine; Head; Health; Image; Location; Manufacturer Name; Medical Imaging; Phase; Process; Reporting; Research; Security; Software Tools; System; Testing; Validation; base; cancer imaging; cloud based; data de-identification; deep learning

This proposal has three main objectives. Firstly, to identify actual and potential PHI/PII locations in DICOM and WSI data - both tags in the header data and coordinates in the image data, including manufacturer specific data - and to deliver a detailed, broad-based landscape analysis report. Secondly, to deliver a robust, validatable software solution that facilitates the development, management, chaining, and execution of medical image data de-identification algorithms. For Phase I, we will implement data de-identification pipelines using as a baseline the data elements identified in the landscape analysis report. Thirdly, to develop, on the same software platform, a set of deep learning-based algorithms to perform image data deidentification. We will facilitate the execution of these algorithms against data both local and remote, in order to obviate any security concerns regarding having to move data to the cloud for de-identification - i.e., we will "bring the algorithms to the data". Our cloud-based solution, EICON REACH (Remote Execution of Algorithms for Clinical Health), provides the underpinning for this proposal. We will enhance its capabilities to address directly the objectives listed. At the end of Phase I, we will provide these capabilities in a fully tested, fully documented, validated solution with full audit trail of user actions and data transformations. EICON REACH can also serve as the basis for a Phase II proposal wherein we would undertake a broader and deeper effort to address the more complex issues related to identification, review and redaction of PHI in the image data.

这项提议有三个主要目标。首先，在DICOM和WSI数据中确定实际和潜在的PHI/PII位置——标头数据中的标签和图像数据中的坐标，包括制造商特定数据——并提供详细的、广泛的景观分析报告。其次，提供一个健壮的、可验证的软件解决方案，促进医学图像数据去识别算法的开发、管理、链接和执行。在第一阶段，我们将使用景观分析报告中确定的数据元素作为基线来实现数据去识别管道。第三，在相同的软件平台上，开发一套基于深度学习的图像数据去识别算法。我们将促进这些算法在本地和远程数据上的执行，以避免必须将数据移动到云中进行去识别的任何安全问题——也就是说，我们将“将算法带到数据中”。我们基于云的解决方案EICON REACH（临床健康算法的远程执行）为这一提议提供了基础。我们将加强其直接处理所列目标的能力。在第一阶段结束时，我们将在一个经过全面测试、文档化和验证的解决方案中提供这些功能，并对用户操作和数据转换进行全面的审计跟踪。EICON REACH也可以作为II期提案的基础，在II期提案中，我们将进行更广泛和更深入的努力，以解决与图像数据中PHI的识别、审查和修订相关的更复杂的问题。