Maxamillion12
Full Member
- Joined
- May 6, 2019
- Messages
- 69
- Reaction score
- 26
Does anyone have an educated guess on this? Have some interest in this, but not if it goes by the wayside. Thoughts?
I was told this was one of the worst questions you could ask somoneshadow a radiologist and ask what they think!
I was told this was one of the worst questions you could ask somone
Because the chances are it is going to go away due to automation/machine learning just like pathology and they are worried about job security?that's right! I think you can figure out why.
Because the chances are it is going to go away due to automation/machine learning just like pathology and they are worried about job security?
Exactly my point...if that's what you think, then I suggest not pursuing radiology residency!
I actually worked at a rural hospital where our on call neurologist was an Australian neurologist who just teleconferences in on a Robot he could drive around. The globalization of medicine is crazy and I think awesome.You will always want a pair of eyes to confirm anything that is computer generated. That said, I think the main problem for American radiologists is that that task can easily be outsourced to international radiologists at a much cheaper price.
Where’s the human compassion though?!I actually worked at a rural hospital where our on call neurologist was an Australian neurologist who just teleconferences in on a Robot he could drive around. The globalization of medicine is crazy and I think awesome.
I mean, if you are a patient who needs a neurology consult at 2 am in a 13 bed hospital in a county of 6,000 people, are you more concerned about getting a disease treated or receiving service with a smile?Where’s the human compassion though?!
I mean, if you are a patient who needs a neurology consult at 2 am in a 13 bed hospital in a county of 6,000 people, are you more concerned about getting a disease treated or receiving service with a smile?
I honestly have no clue what the scenario would be. I just know what they had. And you are probably correct, I made no comment on where they performed their residency.what kind of a neurological scenario in a 13 bed hospital requires a stat consult at 2am? if it's a stroke or a bleed or status the patient should be shipped out to a hospital where they have a neuroICU or interventional neuro or can push tpa
also in your example a physician must have a license in the state they telemedicine to. the doctor may live in australia, but I am 100% sure they are licensed to practice in the US, and as such completed a residency in the US.
This is the same logic as "Even a professional bus driver gets in to car crashes. As a transit user, I would not trust a driverless car."Even very experienced radiologists (at the top academic medical center where I was treated) missed some IMPORTANT things on my CT scans. As a patient, I would not trust a computer to do this job.
I honestly have no clue what the scenario would be. I just know what they had. And you are probably correct, I made no comment on where they performed their residency.
I guess I just approach this topic from the perspective of being a medical laboratory technician seeing automation in the evidence based side of health care first hand.I don't think that I would trust a driverless car at this point, either... If AI is proven to be superior to experienced radiologists, fine, but I just don't see that happening any time soon. And there are a lot of people (i.e. patients) still alive who are much more resistant to technology than I.
Radiology as a discipline won't be defunct until it's determined who will carry the liability for automated reads. There always has to be someone to sue.Does anyone have an educated guess on this? Have some interest in this, but not if it goes by the wayside. Thoughts?
I guess I just approach this topic from the perspective of being a medical laboratory technician seeing automation in the evidence based side of health care first hand.
5 years ago, state of the art chemistry analyzers ran around 20,000 samples per 50 linear feet per day and required 6 technicians, manually checking samples for hemolysis/lipemia/icteremia etc. and a processing department to sort, centrifuge, aliquot and assign to instruments. Today, state of the art runs 75,000 samples per linear 50 feet per 8 hour shift and requires 3 technicians, has a camera to identify any clotting, hemolysis etc. without human intervention, and an entire processing subunit that you literally dump a thousand unspun tubes on to a hopper where it then spins them, produces internal aliquots and sends them to the appropriate instrument. It can even have onboard QC that it draws from on a preset schedule, follow all internal protocols and so on without human intervention. Literally all the technician does is review the ~2-5% of samples that it cannot handle the issue and then review results to make sure they match the patient clinical picture (as far as a bachelors degree person can tell).
Not to compare the difficulties and nuance of radiology to the medical laboratory, but I am just coming from that perspective where automation in the evidence based aide of healthcare is very much real. It is more difficult to automate slide work ups in pathology or sifting through CT scans screening for abnormalities - but if it is a routine task that has regular rules that can be learned/followed then machines can and will eventually do it. If our current state of IS healthcare tells you anything it is that the system doesn’t really care what the patient wants if it makes healthcare delivery more efficient from a profits perspective.
Also, I am just a premed and my premed ego is probably showing. Any physicians that chime in (particularly any radiologists or collaborators on machine learning in medicine) are more likely providing a more accurate answer than I am and these are the personal opinions I hold based on my experiences with healthcare.
Very true. It is not AI, it is automation of repetitive tasks. Like in that Radiology Today article ^^^ it seems like as it is currently projected for the near future, the automation would simply be increased workflow/efficiency. No diagnosis or recommendations would be made, I assume. But making the workflow more rapid I believe is an inevitability.what you're describing is not an AI interpreting whether lab results fit a clinical picture and then the AI suggests a differential diagnosis based on serum and urine lytes and the elements of the HPI... that's something a hospitalist does. What you're describing is just automation of processing samples. not much AI involved.
What specialty is most safe from automation?If there is one thing I've learned from cognitive psychology, it's that the majority of people are absolutely terrible at predictions. Even well trained professionals and leading figures in their field are not an exception to this. Example, there was a study done based on the US leading economists and close to none of them predicted anything like the housing crash of 08'. Now every economist, including the ones in the study, think it was inevitable. The human ego is fragile.
My point being, take any prediction into the "distant" future with a grain of salt. Though your question is interesting, and I'm huge into AI and plan to learn a few things about coding among other things after MCAT studying.
What I can tell you for sure is that -right now- the technology already exists to do what you just explained. Look up "IBM's Watson" supercomputer. It's shockingly accurate! And it's no secret that there is significant human error in radiology reads. This holds true even by the -same- radiologist, reading the -same-scan. These two ideas come together to make me think that there is potential for this, but it's impossible to give you an accurate timeline.
What I can also tell you for sure is even IF they did make this technology able to streamlined in the next 10 years, you'd be hard pressed to see it in hospitals immediately. There would need to be clinical trials and regulatory approval to validate that it's safe and effective to use. That would take an unknown amount of time. Even than, you'll still need a few radiologist to verify their reads during it's early rollout. I'd say you're safe for the next 10 years. 20-50 years? That's impossible to predict.
Also, don't let these people make you feel stupid for asking these questions. It's a valid concern, even if the timeline isn't. The operators of horse carriage didn't picture themselves getting replaced by the car. Now in hindsight, it seemed inevitable.
Edit: If you're truly interested in this topic, and those like it, as it directly relates to medicine there is a new book called "Deep Medicine, how AI can make healthcare human again" by some cardiologist named Eric Tropol. Directly touches on this topic apparently as well. I heard about it at the end of last year, and it is now available. I also plan on reading it after MCAT lol
Why do you think this?The chances are the same as the likelihood these ridiculous threads will stop popping up.
Going to be more like what?It's not going to go away entirely as a field but the numbers needed will probably be reduced. For example you could probably safely automate a lot of routine stuff and leave only the gray areas or complex cases to the fully trained radiologists.
I mean look to anesthesia and the rise of CRNAs. Anesthesiology isn't gone, but the numbers needed to staff certain floors has gone way down now that you can just have one gas doc supervising a dozen CRNA rooms each doing routine GI scoping procedures.
Pathology same idea, you aren't going to fully replace the docs but surely a huge part of the routine biopsy reads can be shifted to machines, or less trained persons assisted by machines.
I personally wouldn't worry about 5-10 years from now, but our careers are going to be more like 30+ years. I'll be absolutely shocked if rads and path resemble their current selves in the year 2050.
also, sorry to OP for me being short.... this question of AI in radiology has been rehashed in popular media a lot. Radiology is not going away and a physician will always read the scans in the near future. AI is pretty lame in medicine. Medicine is not that evidence based or really algorithm based. nothing in the near future will replace the clinical gestalt of a physician. that's why pathology slides are not read by AI and will never be, even though it's probably easier to train pattern recognition of an AI on histology slides. the AI isn't even that good at reading EKGs, what's why a physician signs off on them. I wouldn't worry about it.
Also, I am just a premed and my premed ego is probably showing.
But aren't the preclinical years all memorization too? Couldn't all doctors be replaced? One thing that strikes me is that despite all of this innovation, we still haven't replaced things like fast food workers, cashiers to the full degreeThis. AI can't even read basic EKGs, and I can tell you that a basic chest film is a lot more difficult than a basic EKG. Rads isn't going anywhere anytime soon.
Yes.
But aren't the preclinical years all memorization too? Couldn't all doctors be replaced?
I have. Can't medicine be broken down into you have X pain -> So it has to be Y location. And X is not associated with W or Z, so this has to be the differential diagnosisSuggest that you shadow a physician and see how clinical decision is made in real time.
80% of patients cant even give me a reliable history to do what you describe lol. That's why it takes a really long time to become an attending, because you're learning how to exercise appropriate clinical judgement based on limited or potentially false information. I suggest you shadow in the ED, ICU, critical care and see how decisions are made in real time.I have. Can't medicine be broken down into you have X pain -> So it has to be Y location. And X is not associated with W or Z, so this has to be the differential diagnosis
I have. A machine could easily sift through vitals and figuring out electrolyte abnormalities and piece things together based on probability80% of patients cant even give me a reliable history to do what you describe lol. That's why it takes a really long time to become an attending, because you're learning how to exercise appropriate clinical judgement based on limited or potentially false information. I suggest you shadow in the ED, ICU, critical care and see how decisions are made in real time.
What specialty is most safe from automation?
But aren't the preclinical years all memorization too? Couldn't all doctors be replaced? One thing that strikes me is that despite all of this innovation, we still haven't replaced things like fast food workers, cashiers to the full degree
I have. Can't medicine be broken down into you have X pain -> So it has to be Y location. And X is not associated with W or Z, so this has to be the differential diagnosis
I have. A machine could easily sift through vitals and figuring out electrolyte abnormalities and piece things together based on probability
But aren't the preclinical years all memorization too? Couldn't all doctors be replaced? One thing that strikes me is that despite all of this innovation, we still haven't replaced things like fast food workers, cashiers to the full degree
I have. Can't medicine be broken down into you have X pain -> So it has to be Y location. And X is not associated with W or Z, so this has to be the differential diagnosis
I have. A machine could easily sift through vitals and figuring out electrolyte abnormalities and piece things together based on probability
I have. Can't medicine be broken down into you have X pain -> So it has to be Y location. And X is not associated with W or Z, so this has to be the differential diagnosis
I have. Can't medicine be broken down into you have X pain -> So it has to be Y location. And X is not associated with W or Z, so this has to be the differential diagnosis
I have. A machine could easily sift through vitals and figuring out electrolyte abnormalities and piece things together based on probability
Didn't mean it like this. All I'm saying is that 10 mins ago, everyone was saying that Rads could be replaced, now people are offended by the idea that other specialties can be replaced. The same type of thinking could extend to primary care specialties.Please stop. You have no ****ing clue what you're talking about. It's extremely insulting to all practicing physicians that you would even say this. The amount of thought process that goes into making a decision of what's going on is not simply X Y Z and definitely beyond the comprehension of some premed who did 20? hours of shadowing. So please. Stop talking about things you have absolutely zero understanding about. And no I will not explain anything further because for you to even understand me, you would have to at least had basic physiology and pathophysiology under your belt.
Rads/Path type where “If the Cell looks like this, it is *insert cancer score like the Gleason score for prostate*” or “This shaded area is an inclusion like XXXX and layered images can be produced” are fairly algorithms based. However, even those types of things that will likely be automated to a major degree will merely be automation of the identification process and will inherently be programmed to produce a lot of false positives (so that it catches the largest spread and nothing is missed). And they will not be offering a full differential diagnosis (at least not for the foreseeable future), they will be increasing workflow.Didn't mean it like this. All I'm saying is that 10 mins ago, everyone was saying that Rads could be replaced, now people are offended by the idea that other specialties can be replaced. The same type of thinking could extend to primary care specialties.
Now this makes more sense. Don't know why that guy got so hurtRads/Path type where “If the Cell looks like this, it is *insert cancer score like the Gleason score for prostate*” or “This shaded area is an inclusion like XXXX and layered images can be produced” are fairly algorithms based. However, even those types of things that will likely be automated to a major degree will merely be automation of the identification process and will inherently be programmed to produce a lot of false positives (so that it catches the largest spread and nothing is missed). And they will not be offering a full differential diagnosis (at least not for the foreseeable future), they will be increasing workflow.
Now this makes more sense. Don't know why that guy got so hurt
Most of the docs only attack here without giving an explanation.You premeds should just stop persuing med school and design robo docs. Sounds like you all have this sht figured out.