SynRad-IR-200: A Synthetic Annotated Radiology Report Dataset for Interventional Radiology Triage Benchmarking

SynRad-IR-200 is an open dataset of 200 synthetic radiology reports annotated for interventional radiology (IR) triage, covering IR-positive and IR-negative cases across five clinical domains: Vascular, Tumoral, Infectious, Obstruction-Leaks, and Osteoarticular. The dataset was generated using Gemma-4-31B-Instruct from structured clinical scenarios designed and reviewed by radiologists. Each report is labeled for IR referral indication and urgency. This repository also includes the scenario design spreadsheet and the Python scripts used for prompt generation and report synthesis. SynRad-IR-200 was developed to support the benchmarking of open-source LLMs as automated IR awareness tools in radiology workflows.