A novel platform for synthetic generation and statistical obfuscation of tabular clinical data, simulated images, and machine-generated text

项目总结 数据是一种关键且极具价值的商品，推动着我们的社会发生有意义的变化， 尤其是当它涉及到病人护理和生物医学研究时。目前，机构支付 巨额资金用于增加、恢复和补充他们的数据面板。作为额外的负担， 数据立法和隐私保护法规为形成有效的 商业、临床、研究和教育组织之间的伙伴关系。结果, 目前大约80%的医疗数据不能轻易共享，因为它们包含 个人、受保护或敏感信息，并且在下列情况下仍保持非结构化和未利用状态 都被创造出来了。对平衡技术解决方案的需求日益增长且迫切未得到满足 通过支持灵活的通用功能实现研究和商业组织的利益 分析，同时保证隐私保护。 没有有效的机制来实现 在不冒敏感信息不适当泄露或潜在风险的情况下共享数据 信息内容的退化。目前可用的几种协议和算法用于 建模、处理、询问并最终共享大量敏感数据(例如，数千个 以及具有数千种不同特征的数百万条记录)都具有重要的 其局限性和实际应用仍落后于研究进展。年内两大未得到满足的需求 数据共享行业是i)无法返回原始数据的未识别克隆，以及ii) 缺乏生产部署的可扩展性要求。GrayRain，LLC是一个早期阶段 软件即服务公司开发用于统计混淆和去模糊的新平台 识别敏感的结构化(数字、分类表格数据)和非结构化数据 信息(例如，临床文本、医生/护士笔记和临床图像，如磁共振成像、正电子发射计算机断层扫描)。这个 GrayRain技术的核心是获得专利的新型统计模糊算法DataSifter。这个 在此STTR第一阶段应用中提出的技术将显著增加 确保医疗保健行业及其他领域的数据交易安全，实现数据共享 识别任何敏感信息的可控风险，包括但不限于PHI (个人健康信息)、人口统计信息或社会经济地位。GrayRain的 技术能够产生原始表格数据的未识别克隆，解决了一个主要限制 遇到现有的数据匿名化协议。就可伸缩性而言，它的主要目标是 STTR第一阶段是建立GrayRain准确和高效(Re：可伸缩性)降维的可行性 识别和共享具有可控披露风险的大型复杂电子病历数据存储库 受保护的或个人健康信息。

PROJECT SUMMARY Data is a critical and highly valuable commodity, driving meaningful change in our society, especially when it pertains to patient care and biomedical research. Currently, institutions pay inordinate sums to increase, regain, and complement their data panels. As an extra burden, data legislation and privacy protection regulations introduce barriers to forming effective partnerships between business, clinical, research and educational organizations. As a result, approximately 80% of medical data today can’t be readily shared because they contain personal, protected or sensitive information and remains unstructured and untapped after they are created. There is a growing and urgent unmet need for technology solutions that balance research and commercial organizations interests by supporting flexible general-purpose analytics while guaranteeing privacy protection. There are no effective mechanisms to enable data sharing without either risking inappropriate release of sensitive information or potential degradation of the information content. The currently available few protocols and algorithms for modeling, processing, interrogating, and ultimately sharing large sensitive data (e.g., thousands and millions of records with thousands of heterogeneous features) all share significant limitations and their practical use still lags behind research progress. Two major unmet needs in the data sharing industry are i) the inability to return de-identified clones of the raw data, and ii) lack of scalability requirements of production deployments. GrayRain, LLC is an early-stage Software-as-a-Service company developing a novel platform for statistical obfuscation and de- identification of sensitive structured (numerical, categorical tabular data) and unstructured information (e.g., clinical text, doctors/nurses notes and clinical images, such as MRI, PET). The core of GrayRain’s technology is the novel patented statistical obfuscation algorithm, DataSifter. The technology proposed in this STTR Phase I application will significantly increase the number of secure data transactions in the healthcare sector and beyond, enabling data sharing with fully controllable risk of identification of any sensitive information, including, but not limited to PHI (personal health information), demographic information, or socioeconomic status. GrayRain’s technology is able to produce de-identified clones of raw tabular data, addressing a major limitations encounter across existing data anonymization protocols. As far as scalability, the main goal of this STTR Phase I is to establish feasibility of GrayRain to accurately and efficiently (re: scalability) de- identify and share large-scale complex EHR data repositories with a controlled risk of disclosing protected or personal health information .