Development of a Three-Dimensional Ultrasound Imaging Technique with Auto-Segmentation

The applicability and availability of three-dimensional (3D) musculoskeletal imaging techniques for clinical and research applications are limited by cost and the need for radiation exposure or sedation, particularly in children. This project aims to demonstrate how a machine learning model can generate 3D images from two-dimensional ultrasound imaging, which is quick, cheap, widely available in a range of healthcare settings and radiation-free.

Dr Alex Ireland, Reader in Musculoskeletal Physiology within the Musculoskeletal Science and Sports Medicine Research Centre at Manchester Metropolitan University, and one of the lead researchers on this project, describes the research process, ‘We are training an artificial intelligence (AI) algorithm to recognise different bone and muscle structures using thousands of images. If successful, a doctor, nurse, radiologist should just be able to use a normal ultrasound probe to generate 3D images. Sourcing and labelling the thousands of images required to train the algorithm is a labour-intensive process. We are also using new AI algorithms that have not previously been used in medical setting. Generating 3D images from 2D technology has tried before, although the equipment involved has been rather bulky, making it inconvenient for standard clinical use. Others have tried motion capture technology, which requires huge amounts of processing time and human input.’  Dr Connah Kendrick (Manchester Met), an expert in AI and imaging, and Prof Amaka Offiah (University of Sheffield), a consultant paediatric radiologist, will also bring critical expertise to the project as collaborators.

Alex sees the enormous potential if the project can validate the use of a novel software algorithm to allow automated 3D imaging of bone using a standard clinical ultrasound scanner, ‘Existing MRI and CT scanning technology is expensive and not widely available, especially in developing countries. If we can make the software work with relatively low-cost ultrasound technology, we can put 3D imaging into the hands of a lot more doctors and nurses, including within local GP practices.

‘The potential application to paediatrics is particularly exciting. Using MRI scans for children can be challenging and often requires sedation, even before factoring-in the use of radiation. The technology could also be used on children at the very earliest stages of life when, because their joints haven’t fully mineralised, they are transparent to conventional ultrasound, and for neonatal screening to identify bone conditions, for example malformed hip joints, which is not done routinely in the UK.’