Rüdiger Schmitz

Physician — Physicist — Passionate about #ComputerVision | #Endoscopy | #Gastroenterology | #DigitalMedicine | #ComputationalPathology

[Rüdiger Schmitz @schmitzruediger]

Given the "black box nature" of current AI, updates of medical AI systems or the underlying hardware might lead to extensive re-testing cycles.

What the professional societies can do about that:
https://t.co/nQkGgYT5ni
@Gut_BMJ @ESGE_news @ASGEendoscopy

[Rüdiger Schmitz @schmitzruediger]

@PirnaerDreieck They are counting on you, @DarthPutinKGB

[Rüdiger Schmitz @schmitzruediger]

@alexsworking @dgimev @vision_sharib Plus one or two of the works of your group @alexsworking . Glad that you liked the talk and thanks for coming! :)

P.S. Many thanks for hosting the exciting session, @Rath2Timo (and for inviting me, of course!)

[Alexander Hann @alexsworking]

@schmitzruediger at @dgimev presenting the work of @vision_sharib and showing a possible application scenario where the patient gets his Barrett's as a 3D dataset included in the report.
https://t.co/kaZhWz0XeM

[Annette Hennigs @ABStewardess]

Ladies and Gentlemen, welcome to the post antibiotic era… 😱

[Rüdiger Schmitz @schmitzruediger]

@joostphdrenth @UEGJournal @Clarivate @my_ueg @WileyHealth @GastroRTs @KM_Pawlak Congratulations! 🎉

[Thomas Clausen @DrThomasClausen]

Heute vor 80 Jahren trafen sich fünfzehn deutsche Spitzenbeamte, um die Ermordung des europäischen Judentums zu koordinieren. Zu den historischen Hintergründen der #Wannseekonferenz und der Bedeutung für die Gegenwart hier mein Artikel auf @FNFreiheit: https://t.co/sUT8uGbKhR

[Rüdiger Schmitz @schmitzruediger]

@MazurowskiPhD @dericdesta

[Rüdiger Schmitz @schmitzruediger]

... and will definitely embrace your ideas for our follow-up works, @jansen_marnix :)

[Rüdiger Schmitz @schmitzruediger]

Histopathology-led quality evaluation for #endoscopy/ surgical/ … excisions –– role for #ComputationalPathology and #AI?

Many thanks @jansen_marnix for your insightful comment https://t.co/SOU9ciI14s on our work https://t.co/w8jw0tRcR7 !

@anantm @thomas_rosch @endoscopyjrnl

[Rüdiger Schmitz @schmitzruediger]

@UKEHamburg

[Rüdiger Schmitz @schmitzruediger]

@accessnow @algorithmwatch So perhaps, “#AIact” is just the wrong term for what is intended. It’s more an “automated decision support systems act”, and that’s perfectly fine — We should focus on regulating applications and application scenarios, not so much the underlying technique. 3/@LinaRusch

[Rüdiger Schmitz @schmitzruediger]

@accessnow @algorithmwatch whilst the intention to regulate automate decision support systems in high-risk scenarios also if they use “simpler and rule-based” techniques etc and avoid cutting-edge #ML is more than understandable, this does not make such approaches “#AI”. 2/

[Rüdiger Schmitz @schmitzruediger]

@cabitzaf @FvdSommen @EU_Commission @ki_verband @narges_razavian @LinaRusch @krusehamburg @MarzyehGhassemi @DrLukeOR @AndrewLBeam @aholzin Agreed! Reason why I think we need more effort on thorough testing, incl.: Realistic scenarios, regular benchmarking (if underlying hardware may change etc.), means to test for generalisability (as in the paper you mentioned), outlier detection/dataset similarity measurements,...

[Rüdiger Schmitz @schmitzruediger]

Regulation of #AI in #healthcare ⚕️:

Let's put more focus on benchmarking💯 rather than voodoo "explainability"🧙‍♂️
https://t.co/8PAVFZHrrm

@EU_Commission @ki_verband @FvdSommen @narges_razavian @LinaRusch @krusehamburg -- Thanks to @MarzyehGhassemi @DrLukeOR @AndrewLBeam

[Rüdiger Schmitz @schmitzruediger]

@cabitzaf @EU_Commission @ki_verband @FvdSommen @narges_razavian @LinaRusch @krusehamburg @MarzyehGhassemi @DrLukeOR @AndrewLBeam @aholzin Therefore, on the single case-level, I am afraid that current xAI methods may rather give you a false sense of security. I believe we should be more careful w/ xAI on end-user level and put more effort into benchmarking/ testing scenarios (yes, they are brittle, you are right)

[Rüdiger Schmitz @schmitzruediger]

@cabitzaf @EU_Commission @ki_verband @FvdSommen @narges_razavian @LinaRusch @krusehamburg @MarzyehGhassemi @DrLukeOR @AndrewLBeam @aholzin performes well. It does in fact not even mean that the system used valuable data at all (may still use non-informative single-pixel values etc. in that region). (5/n)