Kimi K3 reads a chest X-ray after full on-device de-identification
MaziyarPanahi · x · 2026-07-20
The author says they got Kimi K3 to read a chest X-ray without seeing the patient’s identity.
They describe an on-device de-identification pipeline from OpenMed that stripped 23 identifiers from the DICOM file, including header tags and pixels, leaving nothing behind. After that, Kimi K3 correctly identified a left-sided whiteout. The post ends by asking whether every scan should be processed this way.
Related event: Kimi K3 Reads Chest X-Rays with On-Device Desensitization(2 posts)→
More from Multimodal
- One prompt, full UGC ad: Kling MCP turns a product idea into ready-to-post video — SimplyAnnisa · 2026-09-11
- A Seedance 2.5 quick-start prompt with GPT Image 2.5 hacks — techhalla · 2026-09-11
- TestingCatalog's Daily AI Brief adds email editions, dishing Meta Muse and GPT-Live-1 rumors — testingcatalog · 2026-09-11
- Reddit user shares WIP AI-generated dark fantasy short film 'Wanderers' — DaWid_Shapiro · 2026-09-11
- Creator makes 2D electro-pop anime music video with just a prompt using MiniMax H3 — Hailuo_AI · 2026-09-11
- Street View to driving footage: GPT Astra fetches images, MiniMax H3 turns them into dashcam video — Hailuo_AI · 2026-09-11