The PACSman, Mike Cannavo.
More often than not, though, especially with panel discussions on AI, I feel like I’m following a decked-out car with a bobblehead in the back window, with everyone nodding up and down agreeing with what is being said by one presenter or another while very little of substance is usually discussed. Just once, I wish I could see people actually have differing opinions.
I also never could understand how some of these clinical studies on AI get published. I just read a study of 2,500 participants in which nearly 600 cases did not mention nodules in the original report. That would be an eye-opener alone if the 24% not found were a concern. Of those initial 24%, only one in five (120 cases or so) were confirmed nodules by a radiologist, and of these fewer than 20 were considered potentially malignant — and ultimately only two nodules likely malignant.
The burning question here is, does improving the findings by 0.08% actually justify the cost of AI technology in this use case? This question is especially crucial because no one knows if those findings identified as potentially or likely malignant were even confirmed until a biopsy and pathology report are both done.
One can argue that saving just one person makes the difference to those people whose finding 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 2,500+ studies in which nodules weren’t mentioned in the initial report that AI purportedly found (almost 600) and then finding out that three out of four of those identified by AI were false positives … well …. what is the cost of that as well? After all, the last time I checked, AI was promoted as a tool to save interpretation time for radiologists, not add to it.
I love seeing positive stories about imaging technology. I got excited when I read a story that started out, “Up to 60% of radiologists have intentions of adopting artificial intelligence tools into clinical practice in the near future.” As the article went on, it said, “… the opinions of those who will inevitably be affected the most by its use — the radiologists — still remain relatively elusive.”
Now, “elusive” is typically a code word for “unsure,” hinting that radiologists probably won’t use the technology. But what about that 60% number? It turns out the study they used surveyed a whopping 66 radiologists. Now, according to the US Bureau of Labor Statistics, there were nearly 30,000 practicing radiologists in the US in 2021. How can one extrapolate a 60% adoption rate from a sample set that equals 0.22% of the overall population? It simply defies logic.
So where is AI imaging technology most likely to be adopted? The answer is simple — where there is the most immediate need. There is a shortage of radiologists worldwide, although the shortage is not nearly as doom and gloom as many would make it. Europe has 13 radiologists per 100,000 population while the UK has just 8.5 per 100,000. Malaysia has 30 radiologists per million or 3 radiologists per 100,000.
It’s not just population density that makes the difference, though, but also the number of studies ordered. That is where the US leads the pack in one area and trails in another. With 11 radiologists per 100,000, the US does well. But France and Germany, for example, have more radiologists per capita. In addition, the more specialized modalities used in the US have longer — and in some cases much longer — read times
Growth of the Medicare population has outpaced the diagnostic radiology (DR) workforce by about 5% from 2012 to 2019. Interestingly, the number of diagnostic radiology trainees entering the workforce increased just 2.5%, compared with a 34% increase in the number of adults. over 65. This is the age group in which most radiology studies are ordered. Compounding matters, 40% of radiologists practicing now are expected to reach retirement age over the next decade.
So what is going to gain acceptance first? In the US, it’s going to continue to be slow growth until reimbursement for using is in effect. In other markets, tuberculosis (TB) screening, COVID-19 screening, and other areas will make embracing AI crucial especially where resources are limited.
Remote digital radiology units on trucks can go to where the patient is to produce the x-ray and then AI can produce a reading in real-time before the patient has left. One new AI model used 165,000 chest x-rays from 22,000 people across 10 countries and tested it against chest x-rays from 1,236 patients from four countries, 17% of whom had active TB. Compared with radiologists, the AI system actually detected TB better with greater sensitivity and specificity, reducing the cost of TB detection by 40% to 80% per patient.
This does not mean that AI is better than radiologists. It’s just that in this selected situation, AI worked well for the application used, especially in developing countries.
AI also has incredible potential to identify the most dangerous potential mutations related to COVID-19, so researchers can get a critical head start on developing protective vaccines. A Swiss team generated a collection of 1 million laboratory-created, mutated spike-protein variants and then trained machine-learning algorithms to flag harmful potential variants that could arise in the future. It is hoped that this knowledge could help create next-generation vaccines and treatments.
This is another area where AI is playing a role in diagnostic imaging although not in the more “traditional” sense of processing imaging data. That is one of the challenges of AI in healthcare — where is it being used and how.
There are literally dozens of applications of AI in healthcare. AI can address everything from enhancing robotic surgery to connecting and taming millions of data points to enhancing the patient experience. That is why one report stated that the AI market is expected to triple in size by 2030 to over $200 billion.
Interestingly, most prognosticators have projected only $500 million in sales for the AI medical imaging marketplace in 2022 and just over $1.2 billion by 2025. That number might sound like a lot, but when you divide it by 200+ vendors with maybe a dozen firms ( if that) currently making money instead of hemorrhaging it … you see the conundrum here.
Where AI goes and how and when it gets there remain question marks, along with most new technologies. Above all, we need to be honest with ourselves regarding answers to these questions and not just nod our heads in agreement with everyone else, hoping whoever does the nodding first is correct.
Michael J. Cannavo is known industry-wide as the PACSman. After several decades as an independent PACS consultant, he worked as both a strategic accounts manager and solutions architect with two major PACS vendors. He has now made it back safely from the dark side and is sharing his observations.
His healthcare consulting services for end users include PACS optimization services, system upgrade and proposal reviews, contract reviews, and other areas. The PACSman is also working with imaging and IT vendors developing market-focused messaging as well as sales training programs. He can be reached at email@example.com or by phone at 407-359-0191.
The comments and observations expressed are those of the author and do not necessarily reflect the opinions of: AuntMinnie.com.
Copyright © 2022 AuntMinnie.com