drwinterpeg

radiologist. teaching society the actual use of AI one post at a time :)

[Alex D. Bibbey, MD @thats_bone]

This happened because a radiologist is working their ass off to get cancer scans read at 11 pm on a Friday before the weekend and it is Federal law that results are immediately available to the patient in the electronic record

[drwinterpeg @drwinterpeg]

@GRomyll Proper auto hanging protocols regardless of series name/institution, auto comparing of target lesions, ambient dictation with auto reporting, triaging of cases would be vastly more valuable than any of these tools. AI companies have the lost the plot

[Romyll Garcia @GRomyll]

As a radiologist, I enjoy these posts. The right-sided chest port already indicates known malignancy and ongoing chemotherapy. It would be far more useful if the tool measured the lesion, incorporated relevant history (including prior treatment), and compared it to the prior exam.

I strongly support Al and find it already highly valuable for written medical work. I also believe it will soon be indispensable for physicians. In radiology, however, I am frustrated by the promotion of nearly useless software by Al leaders and even some radiologists, presented as equivalent to tools like Open Evidence or ChatGPT Astra for physicians.

[Whitfield Lewis, MD 🇦🇬🇺🇸 @whitfieldlewis6]

An innocent neurovascular post has descended into madness. I am now seeing dangerous talking points which can, I think, be collectively described as diagnostic nihilism.

This person says:

Part 1. CTA doesn’t answer questions unless there is an LVO or significant stenosis.

Are you kidding me here? What about the 35-year-old patient who had a TIA and the CTA shows a carotid web? There is no significant stenosis in a carotid web, but clearly, a missed carotid web in a young person with a TIA can be catastrophic. This pathology will result in recurrence of ischemia and ultimately strokes.

What about a free-floating thrombus that may not causes severe stenosis at the carotid? Haven’t you seen such a case? I have. That thrombus continues to break off, which may not cause large, apparent strokes but smaller, subtle infarcts and also TIAs. No CTA for them, huh? And if no CTA, might as well forego the more expensive MRI test.

What about a thrombus sitting in the aortic arch? Good luck appreciating that thrombus with a carotid duplex.

What about that person who reports intermittent dizziness? Good luck finding the cause of this person’s subclavian steal syndrome without vessel imaging. The last time I checked, a CTA would enable you to appreciate the subclavian stenosis proximal to the vert. But hey, it’s just dizziness, so don’t get a CTA unless he has a full-blown stroke.

What about an AVM or dAVF that can cause transient neurological impairment due to hemodynamic compromise to sensitive brain tissue? No stenosis or occlusion there either, but CTA is extremely useful. You know this, right?

Part 2. He says that CTA doesn’t answer questions pertaining to dizziness.

Are you kidding me here? Tell that to Bow Hunter folks whose only complaint may simply be recurrent dizziness with head turning.

Tell that to someone with severe bilateral V4 atherosclerosis with recurrent dizziness due to mild dehydration. I guess we should wait until they get a full-blown stroke before we scan them.

Tell that to the large PICA infarction folks whose chief complaint is, “Doc, I’m just dizzy.”

Nonsense.

Part 3. He also lists headache, diplopia, and mental status change. WHAT??!

This must be April 1st, and I’m being trolled on April Fools’ Day…

What about cerebral venous thrombosis? Wow. What’s the chief complaint here? HEADACHE.

And if that clot gets into the deep sinuses, what will be the presentation here, my friend?? MENTAL STATUS CHANGES!!!

An increased ICP from this pathology will cause what?? NAUSEA AND VOMITING!!

You’re also wrong about diplopia. I guess you don’t believe in or look for a large PComm aneurysm compressing cranial nerve 3, causing diplopia in the process.

Sad!!

@francisdeng @MicieliA_MD @JaysonLavie @northwoods1980 @vsyedavalli

[drwinterpeg @drwinterpeg]

@PotterWorldW Gryffindor is a reverse Lannister crest

[drwinterpeg @drwinterpeg]

@Papa_Heme Zero publications club. Zero research involvement. Zero teaching students. Zero committees and meetings. 100% focus on providing my patients with the best and most up to date care available.

[drwinterpeg @drwinterpeg]

@WillyRontgen @NuanceInc @JoshuaEwellDO @radai @francisdeng @drsxr @northwoods1980 360 is end of life and from what I understand, completely gone next summer. We are in the process of looking for a new vendor.

[drwinterpeg @drwinterpeg]

@JoshuaEwellDO @WillyRontgen @NuanceInc @heacockmd @drsxr @francisdeng Does it still have findings mode or a smart structured reporting?

[Wil @WillyRontgen]

Wow. Two radiologists climbed the Empire State Building.

[Francis Deng, MD @francisdeng]

Isolated nausea/vomiting after iodinated contrast should generally NOT be charted in the allergy history section as a contrast allergy and is NOT an indication for steroid/antihistamije premedication.

Doing so may delay diagnoses without benefit.

Delete bad allergy entries.

[drwinterpeg @drwinterpeg]

@DarthBaneIRL @WillyRontgen BRadiologists know WAY more about what's clinically relevant. We have the authority to stop a study from being performed whether you want it or not. You don't get to order unnecessary studies. We decide what the patient needs as we have the most knowledge.

[drwinterpeg @drwinterpeg]

@Gabe__MD It will never go full AI read, and data points towards that it won't. Not sure where you're getting this info from or what AI product you're forced to shill to make money

[drwinterpeg @drwinterpeg]

@radHQ_net Hi! The sign up button is broken. Getting an error saying I don't have the permission to do that when trying to make an account.

[drwinterpeg @drwinterpeg]

@WillyRontgen @benwhitemd The LLM integration to create structured reports can be helpful. Dictating x-rays is waaaaaay faster than having an AI draft. So they actually created a bottleneck.

[Yannick Buccella MD @YannickBuccella]

Amazing! This is exactly how I imagined how AI could enhance radiology and cancer diagnostics in general!

Detecting pancreatic cancer early will help more patients get life saving surgery, when the tumor is still very small or potentially only in it’s beginning stages.

[Nick Mark MD @nickmmark]

A couple points about why radiologists thriving, not disappearing, with AI:
1️⃣Human + AI is a winning paradigm. I get an immediate AI alert as the scan is done (e.g. stroke suspected), then a couple minutes later a more detailed report/call when the human reads it.
2️⃣Radiologists can read more scans with AI helping. This is an great example of Jevon’s paradox; lower cost of scans leads to more scans being done.
3️⃣Radiologists do a lot more than just read scans; calling to discuss results & further studies is a vital part. I talk to a radiologist at least once almost every time I work in the ICU. I don’t want to talk to a chatbot.

[drwinterpeg @drwinterpeg]

@forgedmedicine @EricTopol Immense number of false positives. Healthcare will implode if tools such as these are implemented. At least it doesn't replace anyone.

[drwinterpeg @drwinterpeg]

@Gabe__MD This study is weak and has terrible sensitivity and specificity despite having a extremely biased retrospective cohort. Ultimately this will lead to drowning the system rather than any improvement in societal quality of life.

[drwinterpeg @drwinterpeg]

@buildthefutures @JimPethokoukis Simply wrong. AI is not even close to replacing what a radiologist does, and in fact is not being trained to in any model. It is there to assist, not to replace.