Researchers have introduced CardioVascBench, a new benchmark designed to evaluate the performance of models in generating prioritized diagnoses from real-world Chinese cardiovascular patient records. This benchmark includes 3,262 de-identified cases, with diagnoses standardized to ICD-10 and linked to clinical narratives and textual ECG reports. An evaluation of 36 systems revealed that higher accuracy in identifying the primary diagnosis does not always correlate with better list precision against discharge references. Additionally, the study presents CardioKG-Agent, a knowledge-enhanced baseline that, when combined with supervised fine-tuning, achieved a 73.5% higher mean composite score compared to fine-tuning alone. AI
IMPACT Establishes a new evaluation standard for diagnostic AI, potentially improving accuracy and reliability in clinical settings.
RANK_REASON The item describes a new benchmark and a baseline model for a specific research task, fitting the 'research' bucket. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Hugging Face Daily Papers →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
