Gadgetron Global Network and Artificial Intelligence Powered Cardiac Imaging

点击翻译按钮获取中文摘要

Past year saw new development in AI applications and data curation. Following applications had been developed and deployed to clinical scanners: . Cardiac perfusion mapping with AI based segmentation and reporting . Cardiac cine imaging with AI based contouring, for retro-gated and free-breathing imaging . AI based cine strain mapping and reporting . Ischemic disease classification using shallow neural nets and reporting on MR scanner . Deep neural net based image enhancement . Deep learning based MR reconstruction More applications are under development: . Lung Cyst detection and auto reporting from MRI . Cardiac function quantification and reporting from cine imaging using AI . Fast cardiac MR imaging and reporting with free-breathing data acquisition - 5 patients per hour protocols for non-contrast CMR study, with full reporting for functional parameters using AI - 3 patients per hour protocols for contrast study, with full reporting of functional parameters, perfusion flow mapping and LGE findings - Clinical evidences collection for values generated by AI powered CMR Is it possible to double the CMR patient throughput with the help of AI, including both faster imaging and faster reporting? . AI powered body and organ fat quantification from free-breathing fat-water MRI Besides developing more AI applications, future work should focus more on AI strategy, infrastructure and evidences for clinical value: . AI strategy: how to develop a system to keep producing AI applications? How to ensure the developed applications have clinical value, besides making publications? How to ensure the involvement of clinical collaborators to guide AI development? . Infrastructure: to support curation of significant amount of data with different sources/types/formats, to support curation of labelled data, to support iterative training/testing/deployment, to ensure ethical compliance and enable international collaboration . Evidences: methods and tools to collect proof that AI applications actually generate values as a part of clinical workflow, e.g. better imaging efficiency (e.g. imaging more patients per day) and better or novel diagnosis. We want to show AI is more than just fancy research.