As an imaging modality to monitor pregnancy, ultrasound has many advantages: it is relatively cheap, portable and gives real time information. The quality, clarity and resulting diagnostic power of the images acquired are highly dependent on the expertise of the person undertaking the exam, usually following a standard protocol. Use of ultrasound by a non-expert is challenging and often results in poor image quality, which has the potential to cause misdiagnosis.
In collaboration with a University of Oxford spin-out business, Intelligent Ultrasound, we have successfully developed novel, fully automated fetal ultrasound screening auditing tools based on machine-learning. These tools allow screening ultrasound scans to be monitored for quality and adherence to protocol. They not only provide a way to assure quality of imaging services, but can also help to identify training needs, and enhance efficiency.
This is a new partnership project between the university and Oxford University Hospitals NHS Foundation Trust and will investigate the extension of the concept: from supporting screening, to diagnostic reporting. This is not easy, because diagnostic criteria are usually less well-defined and more complex than criteria for screening. We will create a machine-learning based tool for assessing the quality of gynaecological diagnostic ultrasound image reporting. Current manual processes will be the starting point and help define the clinical diagnostic criteria, but machine-learning methods will be developed for to automate audit (manual audit is a time consuming and onerous task); this will use clinical ultrasound data as ground truth, information provided by direct observation rather than inference. This will form the groundwork for evaluation in a real world setting with our commercial partners.
Alison Noble, Aris Papageorghiou
Department of Engineering Science and Nuffield Department of Obstetrics and Gynaecology