REG (REport Generation in pathology using Pan-Asia Giga-pixel WSIs)

REG2025 Challenge

This project focuses on advancing automated pathology report generation using vision-language foundation models.
It addresses the limitations of traditional NLP metrics (e.g., BLEU, METEOR, ROUGE) by emphasizing clinically relevant evaluation.
The initiative includes standardized datasets, expert comparisons, and medical-domain-specific metrics to assess model performance.
It also explores the integration of generated reports into diagnostic workflows with clinical feedback.
To support fairness and generalizability, the challenge dataset comprises ~20,500 cases from six medical centers in Korea, Japan, India, Turkey, and Germany, promoting multicultural and multiethnic medical AI development.