PACSman man, Mike Canavu.
Oftentimes, though, especially with panel discussions on artificial intelligence, I feel like I’m following a car with a mottled roof in the rear window, as everyone nodded their heads up and down in agreement with what one presenter or another was saying while the discussion was usually done. Very little substance. Just once, I wish I could see people already have different opinions.
Nor could I ever understand how some of these clinical studies on AI would be published. I just read a study of 2,500 participants in which nearly 600 cases of nodules were not mentioned in the original report. That would be surprising alone if the 24% that were not found were a concern. Of these first 24%, only one in five (120 or so) was confirmed by a radiologist, and fewer than 20 of these nodules were considered potentially malignant – and ultimately only two of the nodules were considered likely malignant .
The burning question here is, does the 0.08% improvement in results really justify the cost of the AI technology in this use case? This question is particularly important because no one knows whether findings identified as potentially malignant or potentially malignant have been confirmed until a biopsy and pathology report has been made.
One could argue that saving just one person makes all the difference to those people whose discoveries would have been missed without the technology, but is it worth the time and cost? When you factor in the cost of reviewing 24% of the more than 2,500 studies where nodules were mentioned in the initial report that the AI allegedly found (about 600) and then finding out that three out of four of those identified by the AI were false positives… well …. What is the cost of that too? After all, the last time I checked, AI was touted as a tool to save translation time for radiologists, not add to it.
I love seeing positive stories about imaging technology. I was excited when I read a story that began, “Up to 60% of radiologists have intentions to adopt AI tools into clinical practice in the near future.” While the article was going on, she said, “…the opinions of those who would inevitably be affected most by its use – radiologists – remain relatively elusive.”
Now, “out of reach” is usually a code word for “not sure,” indicating that radiologists probably wouldn’t use the technique. But what about that 60% figure? It turns out that the study they used polled 66 radiologists. Now, according to the US Bureau of Labor Statistics, there were approximately 30,000 practicing radiologists in the US in 2021. How can an adoption rate of 60% be extrapolated from a sample set equal to 0.22% of the total population? It simply defies logic.
So where is AI imaging technology likely to be adopted? The answer is simple – where there is an urgent need. There is a shortage of radiologists around the world, although this shortage is not nearly as dismal as many might make it. In Europe there are 13 radiologists per 100,000 inhabitants while in the United Kingdom there are 8.5 per 100,000 inhabitants. Malaysia has 30 radiologists per million or 3 radiologists per 100,000.
It’s not just population density that makes the difference, but also the number of studies required. This is where the United States leads the group in one area and moves in another. With 11 radiologists per 100,000, the United States does well. But France and Germany, for example, have more radiologists per capita. In addition, the more specialized methods used in the United States have longer read times – and in some cases much longer
Medicare population growth outpaced the diagnostic radiology (DR) workforce by about 5% from 2012 to 2019. Interestingly, the number of diagnostic radiology trainees entering the workforce increased by only 2.5%, compared to an increase of 34 % in the adult population. Over 65. This is the age group in which most radiology studies are requested. Complicating matters further, 40% of radiologists now practicing are expected to reach retirement age within the next decade.
So what will get accepted first? In the US, slow growth will continue until use is paid for. In other markets, tuberculosis (TB) screening, COVID-19 screening and other areas will make AI adoption critical especially when resources are limited.
Remote digital radiology units in the trucks can go to where the patient will produce the x-rays, and then the AI can produce a real-time reading before the patient leaves. A new artificial intelligence model used 165,000 chest x-rays from 22,000 people in 10 countries and tested them against chest x-rays from 1,236 patients from four countries, 17% of whom had active tuberculosis. Compared to radiologists, the AI system detected tuberculosis better with greater sensitivity and specificity, reducing the cost of tuberculosis detection by 40% to 80% per patient.
This does not mean that AI is better than radiologists. It is only in this chosen case that AI works well for the application in use, especially in developing countries.
AI also has amazing potential to identify the most dangerous potential mutations related to COVID-19, so researchers can get a decisive start in developing preventive vaccines. A Swiss team produced a set of one million lab-created Spike protein variants, then trained machine learning algorithms to identify potentially harmful variants that could emerge in the future. It is hoped that this knowledge will help produce next-generation vaccines and treatments.
This is another area in which AI plays a role in diagnostic imaging albeit not in the “traditional” sense of imaging data processing. This is one of the challenges of AI in healthcare – where and how it is used.
There are dozens of applications of artificial intelligence in healthcare. AI can tackle everything from optimizing robotic surgery to connecting and taming millions of data points to improving the patient experience. This is why one report stated that the AI market will triple by 2030 to more than $200 billion.
Interestingly, most forecasters predicted sales of just $500 million for the AI medical imaging market in 2022 and just over $1.2 billion by 2025. That number might sound like a lot, but when you divide it by over 200 vendors with perhaps dozens of Companies (if that) are currently making money instead of bleeding it… You see the puzzle here.
Where AI goes, how and when it happens, question marks remain, along with most new technologies. Above all, we need to be honest with ourselves about the answers to these questions and not just nod our heads in agreement with anyone else, hoping that whoever nods first is right.
Michael J. Cannavo is known to the industry as PACSman. After several decades as an independent PACS consultant, he worked as a Strategy Account Manager and Solution Architect with two major PACS vendors. He is now safely back from the dark side and is sharing his notes.
Its end-user healthcare advisory services include PACS improvement services, system upgrade, proposal reviews, contract reviews and other areas. PACSman also works with imaging and IT vendors to develop market-focused messages as well as sales training programs. It can be accessed at firstname.lastname@example.org or by phone at 407-359-0191.
Comments and observations expressed are those of the author and do not necessarily reflect the opinions of AuntMinnie.com.
Copyright © 2022 AuntMinnie.com