Exploring Intelligent Healthcare – AI Today and Beyond:
AI in Radiology: From Promise to Invisibility?
Professor Bram van Ginneken, PhD
Cookies: We are sorry we can’t show you this video – as matter of your Cookies preferences. Accept here the Targeting Cookies to watch the video content.
Watch a YouTube video of the full discussion and see more topics in this webinar.
Cookies: We are sorry we can’t show you this video – as matter of your Cookies preferences. Accept here the Targeting Cookies to watch the video content.
Moving from promise to reality and invisibility
Professor Bram van Ginneken, Professor of Medical Image Analysis at Radboud University Medical Center, Nijmegen, in the Netherlands, has been involved in AI since at least 1998 when he started writing an algorithm for detecting tuberculosis (TB) in digital chest X-rays. His algorithm was later adopted by doctors in Zambia because they had a complete lack of readers for X-ray images. “That was even before it was certified,” Professor Ginneken said. “So it was very scary for me, but it also showed me that if you develop something for which there is a big need, people will start using it.”
His algorithm was CE-certified in 2014 and used in over 30 countries to screen around 5,000 X-rays a day. “The World Health Organization concluded in 2016 that every chest X-ray should also be evaluated by a radiologist because there could be other abnormalities besides TB on the X-ray, which is completely true,” reflected Professor van Ginneken.
To Professor van Ginneken’s surprise, more and more organizations started using his software around the world regardless of the WHO recommendation. “In big screening programs, such as in Pakistan, people would line up in front of a bus to get their X-rays taken,” he said. “Then the software would immediately say which of these people had to go to the back of the van, cough up sputum, and get a PCR test to confirm TB. We would then give them treatment immediately.”
The WHO later recommended using computer-aided detection software programs in place of human readers for interpreting digital chest X-rays because many studies showed they worked better than human readers. “This was a very big change and really made AI go from promise to reality,” suggested Professor Ginneken.
His algorithm was CE-certified in 2014 and used in over 30 countries to screen around 5,000 X-rays a day. “The World Health Organization concluded in 2016 that every chest X-ray should also be evaluated by a radiologist because there could be other abnormalities besides TB on the X-ray, which is completely true,” reflected Professor van Ginneken.
To Professor van Ginneken’s surprise, more and more organizations started using his software around the world regardless of the WHO recommendation. “In big screening programs, such as in Pakistan, people would line up in front of a bus to get their X-rays taken,” he said. “Then the software would immediately say which of these people had to go to the back of the van, cough up sputum, and get a PCR test to confirm TB. We would then give them treatment immediately.”
The WHO later recommended using computer-aided detection software programs in place of human readers for interpreting digital chest X-rays because many studies showed they worked better than human readers. “This was a very big change and really made AI go from promise to reality,” suggested Professor Ginneken.
More and more AI software packages are being integrated into products and services. However, customers aren’t always aware of it. For example, Professor Ginneken helped build an AI app for a cheap, commercially available ultrasound probe that connects to a tablet or smartphone. “If you don't sweep the probe over the pregnant woman’s belly correctly, the device beeps and tells you to repeat the procedure,” he explained. “Any midwife can learn to use it in two hours and the app tells them if there are twins, when the expected date of delivery is, and if it's a breach.”
The idea is to get women with high-risk pregnancies to deliver in clinics and not at home, thus, reducing maternal deaths.
The idea is to get women with high-risk pregnancies to deliver in clinics and not at home, thus, reducing maternal deaths.
Figure 2: AI works invisibly behind the scenes to empower this affordable smartphone-based ultrasound solution.
Professor Bram van Ginneken, PhD
Professor of Medical Image Analysis at Radboud University Medical Center, Nijmegen, the Netherlands
Professor of Medical Image Analysis at Radboud University Medical Center, Nijmegen, the Netherlands
“It's doing AI and Deep Learning, but the typical user doesn't see that because it's built into the product,” said Professor van Ginneken.
Professor van Ginneken
“AI may eventually become invisible, but it will still be super useful.”
Overcoming challenges in AI and healthcare
Various Artificial Intelligence issues still need to be resolved. For example, AI arouses suspicion whenever it provides information without sufficient explanation or makes ‘mistakes’. “So it’s important to have a system essentially explain why it is choosing something and then allow you to potentially change the sensitivity and specificity,” maintained Professor Siegel. “And when it makes mistakes, we need procedures to follow up on how to improve the AI,” added Professor van Ginneken,
Bias is another issue the industry is grappling with. “Algorithms that are trained for one specific purpose are often trained on one specific cohort,” noted Mediator, Rich Mather. “And what’s developed on one population doesn't always perform as well on other populations. So it's important to share what population the algorithm was initially trained on,” said Professor Siegel,
Despite current shortcomings, AI offers vast potential for resolving healthcare’s growing workload crisis. Many medical professionals feel they are working on an ‘assembly line’ these days. In particular, it’s no longer practical for radiologists to handle all CT lung cancer screening. “You should delegate a small percentage of the cases to a board of experts to review, but the bulk should be done automatically in screening applications,” Professor van Ginneken asserted. Hopefully, AI solutions will simplify the workflow and vastly increase productivity.
AI also offers great opportunities for reducing disparities in healthcare. “In Africa, they never built a landline phone infrastructure and directly went to mobile phones,”
Professor van Ginneken recalled. “I think they will embrace AI tools, and hopefully refer cases digitally to specialists, to make healthcare more efficient.”
“AI will democratize expertise throughout the world. We should see a significant decrease in disparity with regard to expertise, ” Professor Siegel agreed.
As we start seeing AI applications on app store platforms claiming to interpret chest radiographs for TB or read MRIs of the knee, there will need to be some mechanism for grading performance. Various organizations are trying to establish such initiatives.
“However, the gold standard can be really challenging to set,” said Professor Siegel. “Do you want an application that’s very sensitive and doesn't miss a case, or one that’s very specific and doesn't want to overcall it?”
Professor van Ginneken agreed. “What do we want? I think that’s usually the most important question when you have new technology,” he said. //
Bias is another issue the industry is grappling with. “Algorithms that are trained for one specific purpose are often trained on one specific cohort,” noted Mediator, Rich Mather. “And what’s developed on one population doesn't always perform as well on other populations. So it's important to share what population the algorithm was initially trained on,” said Professor Siegel,
Despite current shortcomings, AI offers vast potential for resolving healthcare’s growing workload crisis. Many medical professionals feel they are working on an ‘assembly line’ these days. In particular, it’s no longer practical for radiologists to handle all CT lung cancer screening. “You should delegate a small percentage of the cases to a board of experts to review, but the bulk should be done automatically in screening applications,” Professor van Ginneken asserted. Hopefully, AI solutions will simplify the workflow and vastly increase productivity.
AI also offers great opportunities for reducing disparities in healthcare. “In Africa, they never built a landline phone infrastructure and directly went to mobile phones,”
Professor van Ginneken recalled. “I think they will embrace AI tools, and hopefully refer cases digitally to specialists, to make healthcare more efficient.”
“AI will democratize expertise throughout the world. We should see a significant decrease in disparity with regard to expertise, ” Professor Siegel agreed.
As we start seeing AI applications on app store platforms claiming to interpret chest radiographs for TB or read MRIs of the knee, there will need to be some mechanism for grading performance. Various organizations are trying to establish such initiatives.
“However, the gold standard can be really challenging to set,” said Professor Siegel. “Do you want an application that’s very sensitive and doesn't miss a case, or one that’s very specific and doesn't want to overcall it?”
Professor van Ginneken agreed. “What do we want? I think that’s usually the most important question when you have new technology,” he said. //
Due to medical device regulatory reasons, not all products/service displayed on this Canon Medical Systems global webpage are available in all countries, regions or markets. Future availability of the products/service cannot also be guaranteed. Please contact your local Canon Medical Systems representative for further details.
