- AI
- Molecular Imaging
- CT
- X-Ray
- Ultrasound
- MRI
- Facility Management
- Mammography
Do Our Own Neurological Speed Limits Pose an Insurmountable Bottleneck in Radiology?
There may be a certain denial about the capability of proposed radiology workforce solutions and technological advances such as AI to enhance our efficiency.
One of my better “follows” on Twitter (or X if you insist) is Brian Roemmele. If you are on that platform at all, I suggest having a look. His stuff almost always piques my interest.
Roemmele recently referenced an article from a few weeks back out of the California Institute of Technology (CalTech) entitled “The Unbearable Slowness of Being: Why Do We Live at 10 Bits/s?”1 I probably read the article more thoroughly than any other scientific write-ups I have seen in the last decade or two.
As you might guess from the title, the piece examines how/why “the information throughput of human behavior” tops out at around 10 bits per second. That number might not seem out of line until you put it into context. The estimated rate at which our sensory systems gather data is on the order of 109 bits per second. That is quite a bottleneck: a hundred million to one ratio.
The article delves into multiple venues in which this disparity can be measured, including expert typists, memorization tasks, and video games to name a few. They didn’t specifically mention diagnostic radiology but enough of the tasks they examined were similar to what we typically do in our line of work. Besides, there is no reason to believe we have somehow stumbled into a profession that magically transcends the limitations of neuroscience.
Not all that long ago, our internal bottleneck wouldn’t have mattered so much in terms of the diagnostic imaging process. Our field’s logistics were such that even an unhurried rad could stay ahead of the workload. Scanners were slower, clinicians ordered less imaging, and technology hadn’t quite gotten to the point where worklists populated faster than they could be cleared. Think even further back, and you can throw in the speed bumps of printed films, alternators, and human transcriptionists.
(It just so happens that around the same time of the CalTech article, a rad wrote about our ever-increasing workload and that crossed my path at around the same time I saw Roemmele’s post.2 They of course didn’t reference one another. Tying them together is what I am here for.)
As everything in our field speeds up around us, we have long since reached the point where human radiologists are regarded as a problematic weak/slow link in the chain. Creative troubleshooting and technological advances have done a lot to increase our efficiency. Over the years I have written about more than a few innovations (both real and wished for). Having changed jobs a couple of times, I have also shared my perspective about how drastically rad groups can differ in terms of tools and infrastructure, which can be helpful or hindering.
Still, there is only so much ground to be gained with that. Sooner or later, one runs up against the ceiling of human capability. Radiologists can only view so many images in a given amount of time, mentally process what they depict, and generate an interpretation. Attempts to exceed that will result in burnout and/or reduced quality. Put in terms of the CalTech article, our 10 bits/s throughput won’t be expanded by nifty new voice-recognition tools or hanging protocols.
The notion that humans have limitations isn’t some brilliant new insight. Knowing that expanding the abilities of individual rads isn’t some bottomless well, folks have long since been pursuing other solutions, many of which involve increasing the number of imaging interpreters. None of these ideas will probably sound new to you, whether it is adding more radiology residency spots, offloading some types of study to other physicians, easing barriers to working across state (or national) borders and even allowing non-physicians to interpret imaging.
Artificial intelligence (AI) is another big one, lest you have been living under a rock and unaware. There is a lot of eagerness for smart machines to become ready for “prime time” so they can swoop in and save us from ourselves, and probably even more eagerness to profit from that by the folks bankrolling the machines. Fortunately, that is counterbalanced (at least thus far) by a strong awareness that prime time isn’t yet upon us.
We are a lot more comfortable with AI in a supportive role. Human rads remain firmly in the driver’s seat, but software is introduced as a helpful tool that may aid in lung-nodule detection, alerts for potential intracranial hemorrhages or large vessel occlusions, and now assistance in generating reports.
Does that begin to nibble around the edges of our 10 bits/s limitation? Well, if we are still going to dutifully look through all the images we are provided and think about each of the findings, maybe not. Actually, if the software presents us with more findings than we would have come up with on our own, that might slow us down. Back in my mammo reading days, for instance, CAD software routinely showered me with false positives. They usually didn’t stop me in my tracks, but each one probably cost me a moment or three.
One possible way to sidestep our neurological speed limit is to add external capability, sort of like adding a lane to a highway that is constantly jammed with traffic. Brain-computer interfaces, a la Elon Musk’s Neuralink, are addressed in the CalTech article, and the authors explain how they don’t expect the 10 bits/s barrier to be surmounted. However, that is just their prediction. There is no way to know for sure until completed prototypes roll out.
References
1. Zheng J, Meister M. The unbearable slowness of being: why do we live at 10 bits/s? arXiv. Available at: https://arxiv.org/abs/2408.10234 . Updated November 15, 2024. Accessed January 6, 2025.
2. McCoubrie P. Moore’s law for radiologists. AuntMinnie. Available at: https://www.auntminnie.com/practice-management/article/15709097/moores-law-for-radiologists . Published November 25, 2024. Accessed January 6, 2025.
The Reading Room Podcast: Emerging Trends in the Radiology Workforce
February 11th 2022Richard Duszak, MD, and Mina Makary, MD, discuss a number of issues, ranging from demographic trends and NPRPs to physician burnout and medical student recruitment, that figure to impact the radiology workforce now and in the near future.
