Gadgetron Global Network and Intelligence Computing: Clinical Imaging Application Development and Software Infrastructure

在过去的2018-2019财年，Gadgetron在人工智能应用开发方面取得了重大进展。使用Gadgetron扫描的部署站点数量和患者数量翻了一番，全球达到40个站点，每年8万例患者。使用Gadgetron生成了大量出版物，我在本报告中列出了一些。更多的报告正在审查和提交。 更重要的是，Gadgetron被开发成一个有效的基础设施，以支持未来的AI研发和临床试验。在所有部署的研究中心中，将大量（每天100例患者）数据发送回NIH。我们策划了这些庞大的数据集，并利用它们开发新的人工智能成像技术。这展示了一种新的模式，不需要向每个站点支付费用，但提供最好的成像和处理服务，以换取高度的互信和巨大的购买和世界上最大的CMR数据集（据我所知）。 下一财年很重要，在我看来是关键的一年，因为人工智能成像在全球范围内以明显的势头起飞，包括CMR。鉴于我们已经取得的成就，我建议以下新目标： .开发人工智能反馈和患者历史界面软件，使Gadgetron成为一个独特的平台来管理成像数据，患者记录和临床反馈。有了这三个关键要素，我们计划进入疾病诊断和自动分析领域（例如预测心脏结局和分类患者是否应该接受干预手术）。 .开发完整的AI驱动CMR分析解决方案，并将其部署到医院用于日常使用，包括电影，LGE，灌注，T1/T2/T2* 映射，脂肪水成像等。这里的动机是完全自动化地提取患者特定的成像信息。这些信息将与患者病史和队列训练的疾病模型一起使用。 .开发用于重大心脏病的MR扫描仪的精确成像。目标是开发患者特定模型，以预测a）患者是否应接受介入手术; B）患者将来是否会发生心脏事件。实现这些的技术路线是：1）自由呼吸CMR成像; 2）AI衍生的成像信息和生物标志物; 3）Gadgetron接收的患者病史和记录; 4）使用队列模型进行预测，其中包含步骤1-3的信息。 为了实现这些目标，也为了不错过这一波人工智能成像的浪潮，我可以建议在即将到来的财政年度增加Gadgetron的资源吗？具体来说， a）聘请一名顶尖的软件系统开发人员。我非常擅长成像应用程序开发和数值算法，但我觉得如果我们能聘请一位专注于反馈系统、病历界面和云计算的优秀软件开发人员，团队将受益匪浅。预期的专业知识将在软件系统，系统间通信，用户界面设计和开发等，主要是非数字软件组件。 B）为Gadgetron建立一个CAN，将其正式确定为一个长期项目，我将领导该项目。其基础是1）Gadgetron每天都在支持NHLBI研究扫描仪和NIH放射学临床扫描仪; 2）Gadgetron每天在全球范围内大量用于扫描真实的患者; 3）由此产生的客户支持不断增加到我几乎无法跟上的水平。为了保持最高的客户满意度（这对我们从部署的站点收集数据和进行人工智能研发至关重要），同时仍然专注于开发新功能和进行新研究，我认为现在是正式确定Gadgetron的好时机，包括其技术开发和客户支持。 请给我一些关于这个建议的反馈，非常感谢！

The past FY of 2018-2019 saw significant advances in Gadgetron for AI application development. The number of deployed sites and number of patients scanned with Gadgetron were doubled, achieving 40 sites globally and 80K patients annual. Lots of publications had been generated using Gadgetron and I listed a few with this report. More are either under review and submitted. More importantly, Gadgetron is developed into an effective infrastructure to support future AI R&D and clinical trials. Among all deployed sites, significant (100 patients per day) amount of data were sent back to NIH. We curated these huge datasets and used them to develop new AI imaging technique. This demonstrated a new pattern which does not require to pay every site, but provide best imaging and processing service, in exchange of high mutual trust and huge buy-in and largest CMR datasets ever accumulated in the world (to my best knowledge). Next FY is important and in my opinion a key year, since the AI imaging is taking off with noticeable momentum around the world, including CMR. Given what we have achieved, I propose following new targets: . Develop AI feedback and patient history interface software, so Gadgetron will be an unique platform to curate imaging data, patient record and clinical feedback. With these three key ingredients, we plan to move into disease diagnosis and automated analysis fields (e.g. to predict cardiac outcome and classify whether a patient should receive intervention procedure). . Develop complete AI powered CMR analysis solution and deploy them to hospitals for daily usage, including cine, LGE, perfusion, T1/T2/T2* mapping, fat water imaging etc. The motivation here is to extract patient specific imaging information with full automation. These info will be used with patient history and cohort trained disease model. . Develop precision imaging on MR scanner for major cardiac disease. The target is to develop a patient specific model to predict a) whether a patient should receive intervention surgery or not; b) whether a patient will have cardiac events down the road. The technical route to achieve these are: 1) free-breathing CMR imaging; 2) AI derived imaging information and biomarkers; 3) Patient history and record received in Gadgetron; 4) Make prediction using cohort model with info from step 1-3. To achieve the goals, and to not miss this big wave of AI powered imaging, may I propose to increase resource for Gadgetron in the incoming FY? In specific, a) Hire a top software system developer. I am very good at imaging application development and numeric algorithms, but I feel the team will benefit a lot more if we can hire a good software developer focusing on feedback system, patient record interface and cloud computing. The expected expertise will be on software system, inter system communication, UI design and development etc, which mainly are for non-numeric software components. b) Establish a CAN for Gadgetron to formalize it as a long-term project, which I will lead. The basis for this is 1) Gadgetron is supporting NHLBI research scanner and NIH Radiology clinical scanners on daily basis; 2) Gadgetron is heavily used world-wide to scan real patients every day; 3) Resulting customer supporting is ever increasing to a level that I can barely keep up. To maintain top customer satisfaction (which is essential for us to collect data from deployed sites and do AI R&D) and still focus on developing new features and conducting new research, I feel it is a good time to formalize Gadgetron, including its technical development and customer support. Please may give me some feedback about this proposal and thank you very much!