Skip to content

Latest commit

 

History

History
32 lines (29 loc) · 1.52 KB

2022-11-7.md

File metadata and controls

32 lines (29 loc) · 1.52 KB

2022-11-7

Agenda

  • Wide variance in CT data (from different scanners?)
    • How can AI be reliable if the underling data differs so much?
    • What about drift?
    • SIIM 2022 AI Monitoring Session
    • https://siim.org/page/
    • Gesund.ai - startup working in this area
    • Ultrasound exploding due to POCUS
    • Since CT values are hounsfield, shouldn't the images have the same HU value for the same patinet with different scanners?
    • Variance in CT Data (Stacy O'Connor). Various things that can affect the same piece of the patient at a time
      • HU 0 = Water (some vendor made it -10?)
      • Upper/Lower bounds are not set
      • Only inner XX cm (32?) is "rated" in terms of the ACR phantom
        • If you are looking at a large patient, the outside is not rated
      • Can get different values for different organs?
        • if you need to compare numbers, must use a phantom to baseline
      • You get different attenuation for the tissue based on the energy (e.g. protocol)
      • Can feed the kernel with raw or reconstructed
    • https://siim.org/page/22m_data_science_summit
    • IHE Management of Acquisition Protocols
    • Some vendor/startup that is adding structured data
    • Wants to use its own ontology
    • Should use radlex
    • flywheel.io claims they have "rules" to structure MR series descriptions with high confidence
      • based on regex?