AI in ultrasound, threat or chance?

01 July 2019

ultrasound AI

Blog by Stevan Stuit – Former diagnostic radiographer and diagnostic medical sonographer

It is everywhere in the news nowadays; the pros and cons of Artificial Intelligence (AI). The influence of AI on economics, labour and our position. One of the most promising areas of health innovation is the application of AI, primarily in medical imaging. The number of publications on AI in radiology has dramatically increased in the last years. Magnetic resonance imaging and computed tomography collectively account for more than 50% of current articles. Nevertheless, only 5% is about ultrasound. What can AI do for ultrasound?

The revolution of ultrasound

In the 80’s ultrasound (US) was the domain of the radiologists, but late eighties, early nineties other specialists like the gynaecologist and cardiologist adopted this imaging technique for their practice. Honesty commands to say that a lot of US exams in the radiology department were, and still are, performed by dedicated US technicians or ultrasonographers. Nevertheless, this diffusion was the beginning of an ongoing process where more and more healthcare professionals started to perform US exams themselves, without the help or supervision of the radiologist. Is US in the Netherlands very common in the midwives practice, nowadays we also see regularly US machines in the GP’s and physiotherapists practice. Even some podiatrists are using this technique. All thanks, in a part, to the unprotected status of US, in contrast to radiation producing equipment like Xray, and the decreased prices of the US scanners. Another trend in the medical environment is the specialisation. We see more and more dedicated medical doctors focusing on a single disease. This specialisation can easily be defended. But does this also apply for the spread of US knowledge?

The adage of the radiology residents is: “The more (US) images you see, the more examinations you report, the better you get”. As an example: where a radiologist, or the ultrasonographer, in a midsize hospital easily performs 40 abdominal US exams per week, a GP –with all respect- carries out this number may be per year. It is impossible for him or her to gain the same experience. You can translate this example to other performers. Given the fact that this distribution is irreversible, there is a call for ongoing help and education.

Artificial Intelligence could be helpful

In the last decade we can see cautious steps of AI, or better ML, machine learning, entering ultrasound. Mostly in the form of automated measurement. In most cases this involves clear visible structures with well-defined edges. Modern automated obstetric measurements are already a bit more complicated for the machine. Also, the latest noise reduction software is in a way done by ML.

The biggest hurdle is the tissue recognition by the US scanner. Researcher have been trying this just as long as my career in US: nearly 30 years. Nevertheless, we are now starting to see AI (deep learning) being deployed to automatically recognise anatomical structures. This image recognition capability is opening the possibility to develop systems which can assist sonographers in real-time through providing diagnostic support. If we can overcome the patient privacy regulations, we can feed a US machine with thousands of labelled images which the machine can learn from. For the machine, this works the same as the earlier mentioned radiologist statement: “The more (US) images you see, the more examinations you report, the better you get”

The art of US will always need the involvement of a physician but AI does provide some opportunities to make this process as efficiently and accurately. For the experienced users, and patients, these intelligent machines could mean shorter scan times and allow users to see more and do more at earlier stages to improve patient outcomes and provide more confident diagnosis. Thereby, the usage of AI in US helps to ensure consistency. This as well helps to provide more confident diagnosis, whether the US is taken on different moments in time or by a different person; this will not influence the outcome. Ultimately, the longer-term outcome is that AI could assist the less experienced user with real time guidance for probe placement in getting the right diagnose and decision-support. Do not only think about the new users, but also count the doctors in the less wealthy countries and in remote areas.

In conclusion, AI will help to achieve the number one goal in healthcare development; higher quality patient care. Therefore, Artificial Intelligence in ultrasound provides great opportunities instead of being a threat.

stevan stuitStevan Stuit, Juli 2019

Former diagnostic radiographer and diagnostic medical sonographer. 15+ years experience in the medical imaging equipment industry with positions in clinical sales, marketing and product management.




  • Artificial intelligence in medical imaging: threat or opportunity? Radiologists again at the forefront of innovation in medicine. Filippo Pesapane, Marina Codari, and Francesco Sardanelli. Eur Radiol Exp. 2018 Dec; 2: 35. Published online 2018 Oct 24
  • Sound the Alarm! Deep Learning & Ultrasound Scans, Shubhang Desai, Stanford AI for Healthcare, Feb 6, 2018
  • Artificial Intelligence within Ultrasound, Stuart Kusta, Signify Research, Dec 11, 2018
  • Artificial Intelligence for Ultrasonography: Japanese Government Policies, Norio Nakata, Ultrasound in Medicine and Biology, 2017Volume 43, Supplement 1, Pages S2–S3
  • How AI is Changing Ultrasounds, Roland Rott, Diagnostic Imaging, November 27, 2018

We want to thank Stevan Stuit for sharing his expertise regarding ultrasound and AI’s impact on it with us. If you have any comments on Stevan’s story, or if you want to share your own story, please contact us via or +31 630076674.