How Machine Vision Applications Are Advancing AI in Medical Imaging

IoT expert shares insights on advancing AI in the medical industry.

Stacey H. Shulman, vice president of the Internet of Things Group (IOTG) and general manager of Health, Life Sciences and Emerging Technologies at Intel Corporation discussed AI in the health care industry at A3’s Vision Week.

Stacey H. Shulman, vice president of the Internet of Things Group (IOTG) and general manager of Health, Life Sciences and Emerging Technologies at Intel Corporation discussed AI in the health care industry at A3’s Vision Week.

The health care industry has always been known for advances, from early vaccines to medical imaging to stem cell therapy and a plethora of others in between. The next step, which has been taken slowly until perhaps it became a side effect of a global pandemic, is artificial intelligence (AI). During A3’s Vision Week, Stacey H. Shulman, vice president of the Internet of Things Group (IOTG) and general manager of Health, Life Sciences and Emerging Technologies at Intel Corporation shared insight into AI’s current and future role in the industry, as well as the challenges in it gaining wider acceptance.

While health care has always been a hot topic, a global pandemic may have helped fuel discussions about it and changed the conversation.

“Overnight, we became a society whose lexicon changed and who started to question just how far behind the health industry was and pondering how quickly it can modernize itself,” Shulman said. “We as consumers have digital access to almost every part of our lives, from banking to nutrition, exercise, education and shopping. I can get my DNA information digitally, but I still can’t get a copy of X-rays, and I still don’t have easy access to medical information. There is a sense that it is an industry that stayed behind. When you talk about AI, it feels like a juxtaposition against an industry that has stopped, and we are talking about a more advanced topic like AI in health care. However, as we look at how much the world has opened to these conversations, we’ve seen the same in the health care industry.”

The Current State of AI

Although AI may seem to be lagging in the industry, there are numerous cases of companies and hospitals embracing it and making it part of the culture. Along with the Mayo Clinic, Cleveland Clinic and U.S. Department of Veteran Affairs, companies like Moderna already embracing AI were ready for the call to action when the pandemic hit.

“They started over 10 years ago on their journey… How do we help the human body create its own medicine?” Shulman said. “They were ready when COVID hit to take the platform they had already tested and use it to quickly build vaccines. Everything they do… AI is incorporated into their decision-making. It is an example of a company that has established a platform and imbedded AI into their culture. Because of that, when an urgent need came up, they were able to make that quick pivot to a therapeutic needed by the industry.”

While other institutions may not have that AI expertise in-house, trends are being set throughout the industry, and perceptions are quickly changing on how crucial AI is to the industry. A survey is done every year in the industry, including the question, “How many years until widespread adoption?” In 2018, the general consensus was that it was going to be a while, likely more than five years. The same question was asked twice in 2020, pre-COVID and after COVID.

(Image courtesy of Intel.)

(Image courtesy of Intel.)

“It happened in the blink of an eye,” Shulman said. “When that hit it, it went from 45 percent saying yes to planning to use AI to 84 percent of companies saying yes, I plan on using AI. It’s not just the AI conversation. They are having a full-digital transformation conversation.”

AI Use Cases

According to Shulman, when you look at the value of AI in health care, you can see the areas where it has the strongest applications. The following are areas that have already been using AI:

  • Patient Positioning:  When someone goes through a CT scan, position matters because it reduces the time the patient is in there. AI is the easiest way to detect where they are.
  • Image Reconstruction: Historically, the quality of images is poor. Part of that was the technology wasn’t updated. AI can be used to fill in the gaps.
  • Automated Disease Screening and Triage: Even for experts, it can take hours to diagnose an X-ray. Using AI ahead of time to do the heavy lifting can take those hours to minutes. 

So, what trends are helping make these AI leaps happen?

GE Healthcare has a first-of-its-kind AI algorithm that is imbedded in the imaging device. It improves the time to analyze, as well as flags critical cases on the device and sends that to the radiologist for immediate triage. Moving AI closer to the patient and device allows for the reduction of data transfer time.

“If you have an enormous X-ray and move it to a device, a centralized device for analysis, that transfer time could take an hour,” Shulman said. “That process had historically not been so automated. Now, moving the AI… As the X-ray is happening in near real time, the analysis can start. It can start looking for things that would be areas of interest for the clinician. It can immediately triage them for the clinician to take a closer look… looking for a collapsed lung, you have to handle them faster than someone with a nodule on their lung.”

NerveTrack is a function that detects and provides information of the location of nerve area in real-time during ultrasound scanning. (Image courtesy of Samsung.)

NerveTrack is a function that detects and provides information of the location of nerve area in real-time during ultrasound scanning. (Image courtesy of Samsung.)

Samsung Medison and Intel have also collaborated on NerveTrack—according to the press release, it is “a real-time nerve tracking ultrasound feature that helps anesthesiologists identify nerves in a patient’s arm to help administer anesthesia quickly and accurately.” Using AI, it can decrease the scanning time by 30 percent.

“If they [anesthesiologists] can identify where the nerve is, they can be more pointed in how they administer medicine,” Shulman said. “It basically still comes back to moving AI to the machine. Let AI happen at the point of interest. Flag those things as a helper to the clinician for better decisions in real-time.”

Another example is Intel’s Huiyi Huiying medical technology (HYHY) and its AI-enabled medical imaging solution that enables the diagnosis of multiple diseases, including COVID-19.  The inference speed in image analysis has been significantly reduced using this technology. According to Intel, inference time decreased by 35 percent for COVID-19. It also has been used for breast cancer detection. The AI-platform resulted in a gain of up to 8.24 times improvement in inference speed with a loss of less than 0.17 percent in accuracy.

The Good, the Bad, the Challenges

With all the good, such as detecting illnesses sooner, gaining faster response times, giving clinicians a better understanding of how to react and more information, there are still many challenges to overcome before widespread adoption truly takes hold.

When it comes to the bad, according to Shulman, “We found a lot of institutions used the same base data. That is a challenge of the industry as well. If you are using the same base data to train a model, it won’t give you the diversity you need. We have to take better care of the types of models, where [we] get them from and really understand the use of the information before we build a model to help us understand the dataset and ask questions.”

Patient privacy and data are two of the other big challenges. Stringent regulations and HIPPA standards, which are in place for a reason, create a hurdle that is difficult to overcome. Since it is an area with strict protections, it slows down the industry by limiting options.

“If we look at it from our own lives, how do we feel about our medical data being used?” Shulman said. “For some of us, if we knew [it] was going to be used [and] anonymized and improving medicine, most would be OK. So, we are looking at better ways to isolate data… keep it in its home and protected but still be able to leverage the insights from it.”

The data itself is also a literally enormous challenge. According to Shulman, 80 percent to 90 percent of it never leaves the hospital, and much of it is image files.

The Future of AI Health Care

To address these challenges, the industry has started to look more at federated learning—the concept of moving the algorithm to the data instead of moving the data.

“Each hospital would train on their own dataset and share the trained model with a federation of other hospitals for a problem, such as COVID,” Shulman said. “We would have regional data and then would have been able to take that and share those models and combine them. The results and accuracy on those combined [are] so much higher than if they used their own validation data.”

Engaging the developer community is another trend. Providers that don’t have in-house AI solutions, or datasets, need the technology to make that happen. Microsoft’s acquisition of Nuance is one way that AI is being brought to providers.

“I feel pretty hopeful about where the industry is going,” Shulman said. “We have to start making baby steps in the industry providing real results and look at the long-term platforms AI should put down so the medical industry can accelerate beyond where they are now… When we establish that, we will be able to respond faster to trends, and quite honestly, save lives in being able to respond quicker.”

Interested in other ways AI is innovating the health care industry? Check out AI Tool Helps Parents Diagnose Children with Behavioral Conditions and Software Can Identify Drug-Resistant Genes in Bacteria That Cause Infectious Diseases.