Close
Ultrasound is the most widely performed medical imaging technique in clinical practice, but is highly operator dependent and requires significant training, skill and experience to perform competently. As a result, both image quality and the reliability of diagnostic information derived from ultrasound images can vary considerably, even between trained operators.
Recent developments in machine learning have demonstrated the possibility of learning from data provided by "domain experts", which can be applied to assist non-expert operators to learn and perform complicated tasks, enabling them to achieve performance levels comparable to those of an expert practitioner. Drawing on these techniques and translating them into a clinical setting, a computer-assisted system, trained using example ultrasound scans performed by an expert operator, can be envisaged that suggests actions that improve a novice operator's ability to navigate. For instance, instructions provided by the system enable them to place the ultrasound probe at the optimal location on the skin surface and orient the probe to obtain high-quality, standardised views of anatomical structures from which key clinical measurements can be obtained.
The overall aims of this project are to design, develop, and test such a system, with a focus on fetal imaging.
Supervisors
Dr Dean Barratt
Mr Raffaele Napolitano
Dr Yipeng Hu