Autonomous cardiac MR Acquisition

Background Cardiac magnetic resonance (CMR) is the gold standard for the diagnosis and evaluation of many diseases, but its long acquisition time makes it a time-consuming and expensive test that is under-utilised in the UK.

Current practice is for a radiographer to prescribe successive imaging planes through which the heart is scanned, but this introduces imprecision and further limits CMR to large centres with relevant expertise.

Aims and Objective We aim to build on our previous work on deep learning and of developing rapid CMR to automate the scanning process, halving acquisition time, removing the need for specialist radiographers, democratising the technology, and increasing precision.

Project plan and methods used We will develop a deep learning algorithm to automate the cardiac MR slice prescription process and implement it directly on the scanner.

A scan-rescan dataset will be created to quantify the algorithm s reproducibility and measure its performance against a cardiac radiographer.

Timelines for delivery This is a 12 month project proposal with training data collection taking three months, six months for algorithm development and deployment, and the final three months spent evaluating it in a clinical setting.

Anticipated impact and dissemination We are confident that we can halve scanning time, reducing the cost of the investigation as well as waiting lists and lowering barriers to use. Precision will be increased, reducing the need for unnecessary follow-up and repeat tests.

Early dissemination will be through collaborating institutions within the NHS and the algorithm will be implemented directly on their scanners.