Using artificial intelligence techniques to do some of the routine jobs in an ultrasound could save hours in a week for sonographers, while maintaining high standards of patient care.
This is according to research at Guy’s and St Thomas’ NHS Foundation Trust and King’s College London. The work was supported by the National Institute for Health Research Guy’s and St Thomas’ BRC and the Wellcome Trust.
For busy sonographers doing up to 12 scans a day, the 7.5 minute-per-scan time saving could add up to one day in a week. The jobs performed by the AI included saving and measuring the images, and creating an automatic report.
The sonographers involved in the study felt the techniques freed them up to concentrate on interpreting images.
The small study involved 23 pregnant women who were due to have their mid-pregnancy scan, at around 18-21 weeks. This scan involves screening for anomalies, and checking that the baby’s internal organs are developing as expected.
The participants had both a regular scan and an AI scan performed by two separate sonographers, so that there would be no impact to care. The two sonographers were randomly assigned to either the AI assisted or the manual scan. In addition, a third sonographer independently compared the images to assess the quality achieved by the two techniques.
Analysis showed that there were no significant differences in the fetal measurements gained by the two techniques. This could have important implications for reducing recognised human error and variation when taking measurements. There was a slight difference in the saved image quality, with the AI tools saving a satisfactory image in 93% of cases, compared to 98% for the manual scan.
Though the scans appeared similar from the patients’ perspective, the sonographers experienced less task switching during the examination. This saved a significant amount of time, which could add up to over an hour a day.
A survey of the sonographers suggested that they felt they could devote even more time and attention to analysing and interpreting the scans when they did not have to complete some of the manual repetitive tasks. The time saving could also reduce work-related repetitive strain injury.
Lead author, Jacqueline Matthew, NIHR Clinical Doctoral Research Fellow at King’s College London, said: “This study demonstrates the feasibility of translating artificial intelligence from the biomedical research domain to a real world clinical application that could have significant benefits for patients and staff in antenatal screening and diagnostic services.”
Professor John Simpson, Professor of Paediatric and Fetal Cardiology at Evelina London, said: “Using artificial intelligence to assist those undertaking fetal anomaly scans has huge potential to improve workflow and quality. This study is an important step, which we are confident will be followed by further advances, such as automated recognition of anomalies to further assist the sonographer.
“The intelligent fetal imaging and diagnosis (iFIND) project team are continuing the development of the AI tools evaluated in this paper. In addition to recognising standard image views and fetal measurements, work is also ongoing to develop tools that can recognise specific anomalies before birth including serious heart conditions. The aim is to further support ultrasound specialists and improve the accuracy of these important examinations.”
