Novel LLM-Based Method for Spectral Material Classification in X-Ray Imaging

Calibrated_Pose_Triplet_Classification_Prompt.docx (36.9 KB)
We’re exploring a new spectral material classification method that uses pose geometry derived from x-ray interactions. The attached prompt uses a 3-angle vector format — called a pose triplet — to compare unknown materials against a reference set and estimate their effective atomic number (Z-effective) using LLM-driven pattern recognition.

This prototype:

  • Uses only 4 energy bins to compute 3 angular values (α_PE, α_CS, Δα)
  • Classifies materials via Euclidean distance in angular space
  • Returns calibrated Z-effective predictions and flags unknowns

We’re inviting others to:

  • Run the prompt in ChatGPT or Gemini
  • Swap in their own [α_PE, α_CS, Δα] triplets
  • Share spectral data (real or simulated) so we can refine and benchmark the system

This is early-stage research from a medical imaging startup (Kairos Sensors) as part of our LLM+physics work. Feedback, critique, or collaboration welcome!