AI is transforming diagnostic imaging

AI feeds off the massive data loads acquired by today’s devices, and then rapidly teaches itself to recognize and characterize disease.

The volume of data that today’s imaging modalities can collect in just a few minutes is vastly greater than the amount of data that a doctor can review in ten times as long. The systems summarize the data in reports, which have become increasingly sophisticated with integration of normative databases, but physicians still need to dedicate a great deal of their own time and judgment to diagnostic imagery. Which of the hundreds or thousands of images are the most important? What do they reveal?

This is where artificial intelligence (AI) is transforming the diagnostic process. AI technologies are now capable of evaluating images and data and presenting those conclusions to physicians. This potential to get greater accuracy in less time is why the AI medical imaging market is expected to reach $2 billion by 2023.1 Today, 17 percent of organizations are using technologies with AI, and an additional 30 percent are planning to do so.2

For Today’s Imaging, Physicians Need AI
The increased power of computers and graphics processors has driven ever-higher performance in all imaging modalities. Not only have imaging systems increased their level of performance, or the speed at which they handle high-volume, high-resolution images, but they have also gained the ability to acquire more and more images faster than ever before.

As a result, radiologists are inundated with data and images to review – a volume so large, in fact, that a recent study estimated they could only assess diagnostic for a typical patient in 3 to 4 seconds in a 8 hours work day3 To utilize all of the data effectively, doctors need diagnostic support.

This is exactly the kind of problem AI is designed to solve. AI makes machines more useful to radiologists by integrating cognitive functions, such as the capacity to compile data, create trends, and build models to improve diagnostic efficacy. As they acquire more data, the machines learn on their own and optimize themselves.

For example, a CT scan of the lungs lasts 3 minutes or less and acquires 5,000 images. Obviously, radiologists cannot go through them, image by image. They need the computer to do a kind of triage, to test the images that are relevant and must be reviewed by physicians in making a diagnosis. This kind of intelligent AI tool offers concrete support for a physician’s work.

Better Complex, Multi-Parameter Diagnosis
On multiple fronts, the world of medicine is progressing three important areas, known as the three P’s: predictive medicine, preventive medicine, and personalized medicine. In AI’s integration with diagnostic technologies, it has the important capability of advancing the three P’s.

For example, in ultrasound technology, AI can predict the severity of a liver disease like nonalcoholic steatohepatitis (NASH). It can play a preventive role in screening, such as determining if a breast lesion is benign or malignant. Finally, AI’s level of data analysis and interpretation means truly personalized care is possible. It understands the differences between one patient’s disease and another’s, so physicians can individualize treatment and continue to utilize the technology to monitor results.

The breast and liver examples are based on ultrasound imaging, but ultrasound and other devices can collect many data parameters. As we acquire all of that data, AI software on a large server can correlate the numbers and derive concrete information for physicians, such as a patient’s probability of having certain diseases. For example, in one of my company’s ultrasound platforms, we are increasing the power 12-fold for AI and adding capabilities to measure tissue viscosity, the amount of fat in tissue, changes in the speed of sound in an ultrasound image, and attenuation, as well as to recognize shape and curve and to automatically measure some data. All of these parameters relate to pathology, and AI uses them to build a multi-parameter picture and deliver conclusions to the physician.

More Data, More Accurate Diagnostic Predictions
Once AI begins making diagnostic predictions, delivering the probability, for example, that a patient’s breast lesion is benign, how much weight does a physician give that probability? Put simply, do we trust that AI is right about something so important?

AI does not replace a physician’s judgment, and unless it has very high, well-documented accuracy, it will simply never be used (and rightly so). The accuracy of AI has to be at least as good as the physician’s, so if the physician is 80% accurate, the machine must be right at least 80% of the time (and save time) to be of any value. And there must be a mechanism to prove this accuracy to physicians.

The best way to make AI more accurate is to accumulate more data, permitting it to learn. This is why cloud-based architecture is so important for AI. When a system can share and combine data with other databases inside and outside the walls of an organization in an anonymized fashion, data builds exponentially, the system learns from the data, and it becomes more precise in its data evaluation. The AI system’s capabilities grow continuously and autonomously.

Automating Protocols, Saving Time
As reimbursements for imaging exams have decreased, equipment costs have remained the same. Only by making imaging faster and more efficient can we make this combination profitable. I mentioned ways that AI saves physicians time interpreting data and contributes to more accurate diagnosis, but the technology also can cut minutes off an imaging exam – minutes that add up very quickly.

For example, AI incorporated in the machines’ user interfaces can automatically tune the images, correcting factors such as gains of attenuation in an ultrasound. Doctors and technicians don’t need to make those adjustments.

Doctors also can make automatic protocols in the system to automatically guide the examination, ensuring that a radiologist does not forget to acquire all the key parameters, including taking new forms of measurement as they’re added to the system. None of this is a burden to the exam – in fact, it is almost transparent to the user. It’s how imaging technologies with AI provide valuable, streamlined support for clinical care, from capture to diagnosis to monitoring treatment.

Jacques Souquet, Ph.D., is Founder and Chief Strategic and Innovation Officer at SuperSonic Imagine, a company specializing in ultrasound medical imaging. He holds 10 patents in the field of acoustic imaging and is the inventor of the multiplane transesophageal echo probe, which is widely used in echocardiography.

1. Harris S. AI in Medical Imaging to Top $2 Billion by 2023. Signify Research. August 2, 2018. Accessed online November 25, 2018:
2. Bresnick J. AI for Imaging Analytics Intrigues Healthcare Orgs, Yet Starts Slow. HealthITAnalytics. February 17, 2018. Accessed online November 25, 2018:
3. McDonald RJ & al from Mayo Clinic The effects of changes in utilization and technological advancements of cross sectional imaging on radiologist workload Acad Radiol. 2015 Sep;22(9):1191-8. doi: 10.1016/j.acra.2015.05.007. Epub 2015 Jul 22

© Copyright ASC COMMUNICATIONS 2019. Interested in LINKING to or REPRINTING this content? View our policies by clicking here.


Top 40 Articles from the Past 6 Months