RAPTOR+ Evaluation Outputs: Ground Truth and Results for Colorectal Cancer Referral Triage

This dataset contains the evaluation outputs generated by RAPTOR+, a grounding-aware vision-language pipeline for colorectal cancer referral triage. Included are ground truth annotation files in JSON format, generative AI model outputs containing structured referral data extracted from colorectal cancer referral forms, and visual grounding results across evaluated model configurations. The dataset was used to benchmark RAPTOR+ against clinical ground truth established by expert clinicians. Associated paper: RAPTOR+: A Visually Grounded Vision-Language Framework to Improve Clinical Trust and Auditability in Automated Cancer Referral Processing, submitted to Cancers (MDPI). Synthetic referral data was curated by clinical experts at University Hospitals Birmingham NHS Foundation Trust.