{"doi":"10.1101/2025.07.13.25331222","title":"A Clinically-Informed Framework for Evaluating Vision-Language Models in Radiology Report Generation: Taxonomy of Errors and Risk-Aware Metric","abstract":"Recent advances in vision-language models (VLMs) have enabled automatic radiology report generation, yet current evaluation methods remain limited to general-purpose NLP metrics or coarse classification-based clinical scores. In this study, we propose a clinically informed evaluation framework for VLM-generated radiology reports that goes beyond traditional performance measures. We define a taxonomy of 12 radiology-specific error types, each annotated with clinical risk levels (low, medium, high) in collaboration with physicians. Using this framework, we conduct a comprehensive error analysis of three representative VLMs, i.e., DeepSeek VL2, CXR-LLaVA, and CheXagent, on 685 gold-standard, expert-annotated MIMIC-CXR cases. We further introduce a risk-aware evaluation metric, the Clinical Risk-weighted Error Score for Text-generation (CREST), to quantify safety impact. Our findings reveal critical model vulnerabilities, common error patterns, and condition-specific risk profiles, offering actionable insights for model development and deployment. This work establishes a safety-centric foundation for evaluating and improving medical report generation models. The source code of our evaluation framework, including CREST computation and error taxonomy analysis, is available at https://github.com/guanharry/VLM-CREST.","journal":"medRxiv","year":2025,"id":555663,"datarank":0.0,"base_score":0.0,"endowment":0.0,"self_citation_contribution":0.0,"citation_network_contribution":0.0,"self_endowment_contribution":0.0,"citer_contribution":0.0,"corpus_percentile":null,"corpus_rank":null,"citation_count":3,"citer_count":0,"citers_with_citation_signal":0,"citers_with_endowment":0,"datacite_reuse_total":0,"is_dataset":false,"is_dataset_confidence":0.9528,"is_data_producer":false,"deposit_databanks":null,"is_oa":true,"file_count":0,"downloads":0,"has_version_chain":false,"published_date":"2025-01-01","fair_score":null,"fair_percentile":null,"algorithm_id":"datarank_citation_only_1hop_v6","ranking_scope":"data_only","authors":[{"id":295498,"name":"Peter C. Hou","orcid":"0000-0002-7726-2901","position":1,"is_corresponding":false},{"id":805224,"name":"Pengyu Hong","orcid":"0000-0002-3177-2754","position":2,"is_corresponding":false},{"id":1454467,"name":"Liqin Wang","orcid":"0009-0002-6389-4403","position":3,"is_corresponding":false},{"id":1454468,"name":"Wenyu Zhang","orcid":"0009-0001-1457-1707","position":4,"is_corresponding":false},{"id":432794,"name":"Xinsong Du","orcid":"0000-0003-3713-3264","position":5,"is_corresponding":false},{"id":1242580,"name":"Zhengyang Zhou","orcid":"0000-0003-4728-7347","position":6,"is_corresponding":false},{"id":1454469,"name":"Li Zhou","orcid":"0000-0002-0300-0394","position":7,"is_corresponding":false},{"id":629655,"name":"Hao Guan","orcid":"0000-0002-8117-4584","position":0,"is_corresponding":true}],"reference_count":15,"raw_metadata":null,"created_at":"2026-07-19T02:55:03.976486Z","pmid":"40791731","pmcid":null,"fwci":null,"citation_percentile":null,"influential_citations":0,"oa_status":null,"license":null,"views":0,"total_file_size_bytes":0,"version_count":0,"fair_f":null,"fair_a":null,"fair_i":null,"fair_r":null,"fair_zscore":null,"fair_rationale":null,"fair_model":null,"fair_agent_version":null,"fair_fulltext_source":null,"fair_has_llm":null,"fair_computed_at":null,"clinical_trials":[],"software_tools":[],"db_accessions":[],"linked_datasets":[],"topics":[]}