复旦大学最新Cell子刊：DeepSeek-R1、GPT-4等大语言模型可增强肺癌筛查的临床决策

Core Insights - Lung cancer is one of the most aggressive and prevalent cancers globally, with an estimated 2.2 million new cases and 1.8 million deaths in 2020, leading to a five-year survival rate of less than 10% due to late-stage diagnosis [2] Group 1: Research Findings - A multi-center benchmarking study evaluated six large language models (LLMs) for clinical decision support in lung cancer screening, revealing that Claude 3 Opus had the highest readability, while GPT-4 achieved the highest clinical accuracy [3][7] - The study involved a cross-sectional analysis of 148 anonymized low-dose computed tomography (LDCT) reports from three medical institutions, assessing the performance of LLMs in providing management recommendations for incidental lung nodules [6][8] - The results indicated that the performance differences among LLMs were not significant across different hospital reports, highlighting their robustness and practicality in various medical environments [7][10] Group 2: Implications for Clinical Practice - The findings suggest that LLMs could enhance clinical decision support in lung cancer screening, particularly in managing incidental findings from LDCT scans, which is a pressing challenge in cancer screening management [6][10] - The study underscores the potential of LLMs to assist outpatient physicians in making timely decisions regarding follow-up interventions or surveillance strategies for lung nodules [5][6]