GaryGaoYao/OrbitalPSI-GPT

This is for the code sharing of our work "A Text-Driven Clinician-in-the-Loop Approach to Accessible Patient-Specific Orbital Implant Design in Low-Resource Settings" # Designing Patient-Specific Orbital Implants with Just Your Words （In Preparation） Code repository for **OrbitalPSI-GPT**, accompanying the work: > **A Text-Driven Clinician-in-the-Loop Approach to Accessible Patient-Specific Orbital Implant Design in Low-Resource Settings.** ## 🎥 Demo Video **Zenodo (Restricted Access):** doi.org Availability (click to expand) - The video is currently under **Restricted Access**. - It will be made **publicly available after the publication** of our work. - The Zenodo DOI remains stable and can be cited at any time. - Researchers wishing to access the video before publication may request access directly via the Zenodo page or email: gary.gaoyao@gmail.com. ## 🧠 System Overview OrbitalPSI-GPT enables surgeons to design patient-specific orbital implants through natural-language instructions, without requiring CAD expertise or proprietary software, particularly in low-resource settings. For more information or technical details, please refer to our paper (to be updated upon publication): ## ✅ Clinician-in-the-loop Workflow （Our highlights） We break the design process into a few simple checkpoints (CP0–CP4). You stay in control from start to finish — using just your words. Workflow (click to expand) 1. **CP0 — Case setup: Tell the system what’s going on** You start by giving a short description of the case (which side, what type of defect, etc.). The system automatically extracts the key information and prepares the case for you. 2. **CP1–CP2 — Anatomy check: Make sure the “map” is correct** The system loads the CT scan and the segmentation results. You quickly look at the 3D view to confirm everything is labelled correctly. 👉 **If you don’t approve it, the system will not generate any implant.** 3. **CP3 — First draft: The system creates an initial implant shape** Using expert-designed templates mapped to the patient’s orbit virtual reconstructions, the system generates a smart “fir …