Risk Tolerance

MD PhD in iPSC immunology

Rich from biotech and mining speculation

[Risk Tolerance @_RiskTolerance]

@1cavemank @gfodor nodularity in specific location vs bogginess vs exquisite tenderness vs general enlargement does tell u info tho. it changes the pretest probabilty for PSA interpretation

i dont deny that time series might be useful - but MJ's US isnt. it will first be via MRIs for many reasons

[Risk Tolerance @_RiskTolerance]

@SteveGarciaMD @justindubinmd okay but MDs send ppl to the ER all the time for asymtpomatic hypertension

most annoying thing ever but doctors do it anyways

[Risk Tolerance @_RiskTolerance]

@DoeSparkle @justindubinmd >person with weird sexual orientation is calling other people annoying hypochondriacs

wew

[Risk Tolerance @_RiskTolerance]

@ICUPA1 @oitixion i literally use claude all the time for my health concerns and im a physician so

[Risk Tolerance @_RiskTolerance]

@oitixion @Benjami13759491 NPs are so bad for the average person, they miss basic diagnoses and drug interactions. if they want to practice medicine, they should pass medical boards

they are good for high IQ healthy people who just want to place an order for what they want (give me my adderall refill)

[Risk Tolerance @_RiskTolerance]

@StephenPiment MJ product is just not good tho

if u really wanna do preventative scans, we have MRI

if we cant train AI to read MRIs with billions of high res scans and assoc diagnoses/biopsies/history, we are not gonna get AI to read shitty low res US scans with no pre-existing database

[Risk Tolerance @_RiskTolerance]

@FertileFitness @StephenPiment i mean ur data shows that the clear majority of attending physicians understand it, whereas medical students and junior residents dont?

im not even simping for doctors i hate them but at least have ur evidence support ur argument

[Risk Tolerance @_RiskTolerance]

@subaxiomat most frustrating thing is libtarded SF tech bros shitting on doctors for midjourney skepticism but these same faggy tech bros are all TRUST THE SCIENCE when it comes to vaccines and covid lockdowns

[Risk Tolerance @_RiskTolerance]

if preventative scans could work in theory, it would make more sense for them to work on the highest resolution images (MRIs, even thin cut MRIs) with billions of existing images to train on with known associated disease states

it would make way more sense to do preventative MRI scans on a cohort of 10,000 ppl, and if those showed benefit to expand preventative scans. the issue is they dont currently work, AI cant reliably read MRIs with glaringly obvious pathology (much less subtle incipient lesions), and this US has fundamental physical limits to it

if it would work, i would expect MRIs to work way way way first, and then for ppl to train US by doing a MRI and US together and using the MRI data to train the US read

but MJ's promo material is way overhyped bordering on fraud

[Risk Tolerance @_RiskTolerance]

@1cavemank @gfodor thats my point, FYI

literally taking something up your ass for no benefit (but its more data!)

[Risk Tolerance @_RiskTolerance]

@cruthaifios @ByrneHobart im not against training AI to replace radiologists. I'd love it

but i dont like MJ hype. we have 100s of millions of MRI scans with symptomatic history and subsequent biopsy/confirmatory diagnosis. it makes far more sense to train on that than US which has many limitations

[Risk Tolerance @_RiskTolerance]

?

A thin slice full body MRI is superior to prototype continuous US?

I eagerly await the day AI can read a MRI better than a radiologist. its years away and we have billions of scans to train on, with history+indication to correlate the scans with. Asymptomatic scans done via novel tech with no history or symptoms to correlate with will be much harder to train models on, especially since we dont have 100s of millions of scans ready to go

[Risk Tolerance @_RiskTolerance]

@NuggetCapital i guarantee you all the institutional spec money has long since been washed out trying to touch oil

Open interest at historic lows for oil

[Risk Tolerance @_RiskTolerance]

@FrenlyOfficer @elon @Nikita isnt this more of an Officer Frenkly type post

[Risk Tolerance @_RiskTolerance]

@JOSEPH45075332 @quasistable @AmandaAskell RE: recommending serial imaging over time

do u change the frequency from gen pop guidelines (ie. q2 yrs 40-49 to q1y if lesion with benign char detected)

or is it just maintain normal mammography schedule

[Risk Tolerance @_RiskTolerance]

I agree and think its an exercise that is near futile and techbros are gonna learn there are limits to technology, as the finance bros will learn that excel models don't always and everywhere forecast reality

my understanding is there is never a use for serial breast imaging currently. its just change modality and/or proceed to biopsy.

[Risk Tolerance @_RiskTolerance]

>they don't differ morphologically

i think the hope is that while this is true to the human mammogram reader, AI will be able to detect shades of gray and other patterns imperceptible to the human eye and hugely improve screening.

but I agree - if u cant do it on something with well defined criteria, and literal tens of millions of preventative scans, and post scan biopsy data, then it is going to be a hugely losing battle for undefined amorphous abdominal blobs via low res US

i think if AI can truly deliver imaging alpha, breast would be the most likely domain for it to appear first

[Risk Tolerance @_RiskTolerance]

@JOSEPH45075332 @quasistable @AmandaAskell we might be coming up to a point where we learn via mass (non-IRB approved) experimentation

do 100,000 unindicated abdominal mass biopsies done and we can learn which characteristics (density, size, calcification, shape, etc) would actually make incidentaloma biopsy worthwhile

[Risk Tolerance @_RiskTolerance]

@JOSEPH45075332 @quasistable @AmandaAskell i just dont like the modality tbh

I think serial MRIs make way more sense if we rlly can solve imaging via brute force compute. MJ trying to solve an old problem with an inferior tech cuz its maybe quicker/cheaper

biopsies also have complications/adhesions/seeding problems