Anyone familiar with military technology knows there are many instances of it being adopted outside of martial matters.
From Silly Putty to ultrasound to satellite imaging, there’s an established history to back the claim. In terms of the latter on the list, there’s an extra step occurring — it’s not a close translation of satellite imaging to a use like road mapping for example. Instead, it’s adapting satellite image theory and pairing it with operational AI to help with healthcare’s big data challenge. To learn more, HealthCare Business News spoke with Avi Veidman, CEO of Nucleai, an Israeli startup seeking to help pathologists find and battle cancer.
Veidman has a background in engineering and AI. He served 20 years in the Israel Defense Forces’ Technology Unit, ultimately serving as head of AI in the Military Intelligence Corps. When he retired in 2017, he decided to use his knowledge and the networking connections gained when developing systems for the Israeli military to find solutions in healthcare.
The work both for the military and healthcare entailed the development of algorithms and systems. In healthcare, the intended use is in pathology and precision medicine. To understand how the medical applications work, one needs to simply consider what satellites do. “When you’re looking at satellite images, you’re looking for very concealed objects which are tiny,” Veidman explains. “I thought, if you can detect concealed missiles or tanks, why can’t you use technology based on a similar premise to detect cancer?”
Veidman explains that combining various sources of information, including satellite images, provides complete intelligence and insights. That’s the same approach in healthcare, paired with clinical information from genomics and labs. “In that sense, there are patterns quite similar in the application point of view,” he says.
This could be a boon to the healthcare system. In radiology there are various sections not being scanned. In pathology, not every pixel is being investigated, which means useful information may be missed.
Recent studies have indicated that while medical imaging accounts for 90 percent of all healthcare data, less than 3 percent of that information is analyzed or used. That gap between collection and usage is part of what Veidman and his team are trying to address. “Any data that was collected was done so for a reason,” he says. “However, if you have a biopsy where you’re imaging the human body, going over all the pixels is not feasible. You could not take into account all that information and that’s where computers come in. They are good at providing much more thorough coverage of those images. They are very proficient at looking into the areas where something is suspect.”
The algorithm Veidman and his team uses reviews the image in the highest resolution available. In the case of the pathologist, the AI could help to not only detect the areas they might have missed, but make the work more efficient.
While there is a lot of promise for AI to tackle the data, there is still work to do in order to close the information gap. “I’ll give an example with a scanning algorithm in the case of prostate cancer. When someone detects what might be prostate cancer, another pathologist will review that case,” Veidman says. “In 30 to 40 percent of instances, the cases are benign and no one else is going and reviewing those benign cases. So we have an algorithm that reviews all the cases and helps to determine if there were any discrepancies or a misdiagnosis, and then it raises the flag if necessary to let the pathologist know to review those particular images again.”
The more information fed into AI, the more powerful the tool. Veidman acknowledges this, and that was why he worked to develop partnerships with the biggest hospitals in Israel. Those relationships provide a pool of 20 million slides to access — critical for developing the algorithms that deliver the accuracy and specificity needed.
Veidman says there are pathologists reviewing those slides and marking them — benign, malignant, etc. — in order to train the AI. For those in the field doing imaging of patients, the technology largely runs in the background, with little impact on their work.
While he’s optimistic on the convergence of AI, satellite imaging theory and medical imaging, he’s less confident on when we might see a high level of adoption. “It’s a prediction that involves human beings, so it’s probably the hardest. We could take a look at the past to get some examples though. What was the time it took all the systems to get new technology involving IT in the past? Like EHR for example? Maybe it was 10 or 15 years, maybe still ongoing in some places,” he says. “I think it will continue to grow steadily, especially in radiology and ophthalmology and very soon in pathology. It’s going to be a very long journey because of the technology, people reluctant to make the change to something new, and of course, regulations. But, the train is out of the station.”