Today's Nanox AI event showed a sample image from "nanox.Arc demo unit" that was not submitted for regulatory approval.
Can someone explain what I may be seeing here - aren't there too many bones? Was this early detection of an alien lifeform living inside a phantom human hand?
Update November 5, 2021: The recording of Nanox AI event posted on Youtube has been made private and cannot be viewed anymore. I wonder why. Time to notify the FDA about the alien discovery by a device about to be cleared?
Update November 9, 2021: Recording is viewable again, under new URL.
Update November 14, 2021: I was asked on Youtube how I can be sure that the June submission for FDA clearance was fraudulent. Dr. Orit, the Chief Medical Officer, admitted during the webcast that 1) the Nanox.ARC device is currently under development (49m:44s), and the "images" [including the disastrous tomosynthesis slab of the "alien lifeform" hand] are currently not usable for diagnostic purposes - "it is still being developed" (52m:07s). Therefore, the device submitted for clearance in June was not ready to be marketed or tested, which is fraud.
Looks like a double exposure or an image from the multi-tube arc with a few tubes running at once. Eventually they may get to tomosynthesis!
ReplyDeleteDouble or Quintuple? How few? Are you saying the multi-source "nanox.Arc demo unit" is actually a two-source device? I think we better stick with the alien lifeform hypothesis, for Ran's sake.
DeleteIts a multi layer image. Computers understand it. Used as AI learning input.
ReplyDeleteDumbass.
Yes, in the Andromeda Galaxy, computers understand it and use it as an AI learning input. In our galaxy, it is called fraud.
DeleteSpecifically, two things are clear. First, Nanox Arc latest hardware, whatever that is, is creating severely underexposed and blurry projections, due to slow and weak x-ray source and detector combined with short SID and poor positioning. Second, the noise overwhelms the tomosynthesis algorithms used by Nanox and so they are unable to do a decent reconstruction and introduce severe artifacts ("shadows," "ghost layers," "double vision," etc.). In other words, the best images that Nanox can make today are completely useless for diagnostic purposes or for training AI. The FDA submission in June must have been fraudulent.
Multiple images superimposed would be far less useful , both to human and AI , than individual image planes in parallel.
Delete@Jeremy Since neural networks are optimizers (curve fitters), just like the iterative tomosynthesis reconstruction algorithms, the raw 45 projections of the hand obtained during the scan SHOULD be more useful for AI. But if the projections were any good, and there was some hope, someone at Zebra should have spoken up. Instead, Zebra's Dr. Wimpfheimer had to present an alien image. This leads me to conclude that the individual raw images (that
Deleteis, the "single-source" projections at different angles from 5 sources at 9 tilts) are themselves unusable, except maybe a marginally useful underexposed one taken at perfectly perpendicular to the detector - a plain x-ray.
Hi Bro,
ReplyDeleteDo you have any comments about NNOX 3D images comparing with regular 2D Xray images on the market right now? Do you think it is real images from NNOX ARC or they just faked images from their database? How come the images look so real? Please write an article explaining about how they can get these images from their fake tubes and fake machine . Thanks bro!
Yes, I do, Sis. Regular 2D Xray images are useful in many situations - please consult ACR Appropriateness Criteria
Deletefor a comprehensive list ( https://www.acr.org/Clinical-Resources/ACR-Appropriateness-Criteria ). Real tomosynthesis images (other than breast) are NEVER "usually appropriate." And Nanox 3D images are completely useless, based on what we have seen so far, and are much worse, obviously, than whatever tomosynthesis images you can get today with the cheapest equipment (a tomo is just a $30,000 software upgrade of a regular GE machine, for example, but very few professionals request it - tomo has less than 1% market share of medical diagnostic equipment using x-rays). I don't know yet how Nanox generated the bad images in the RSNA 2021 infomercial, but the images definitely did not come from the Nanox.ARC scans shown there (easy to prove using timestamps, etc).