Ding Zongren, Zeng Jianxing, Fang Guoxu, Guo Pengfei, Zhou Weiping, Zeng Yongyi
Department of Hepatopancreatobiliary Surgery, Mengchao Hepatobiliary Hospital of Fujian Medical University, Jintang Road 66, Fuzhou, China.
The Big Data Institute of Southeast Hepatobiliary Health Information, Mengchao Hepatobiliary Hospital of Fujian Medical University, Fuzhou, China.
Sci Rep. 2025 Jul 29;15(1):27549. doi: 10.1038/s41598-025-08502-4.
Primary liver cancer is the sixth most commonly diagnosed cancer globally and the third leading cause of cancer-related deaths. Hepatocellular carcinoma (HCC) is the most common type of primary liver cancer, and microvascular invasion (MVI) is a significant risk factor affecting postoperative prognosis in HCC. However, accurately predicting MVI preoperatively remains a challenge. This study aims to evaluate the application of large language models (LLMs), specifically ChatGPT 4o, in predicting MVI in HCC and to compare its performance with traditional clinical models. In this retrospective study, 300 HCC patients who underwent curative liver resection between June 2018 and December 2018 were selected at two centers. The collected clinical data included age, gender, HBV infection, liver cirrhosis, AFP levels, and more. ChatGPT 4o were used to process the clinical data of the patients and predict MVI. Subsequently, the predictive results of the ChatGPT 4o were compared with machine learning models, the ROC curves were plotted, and AUC was calculated. The results showed that the AUC of the ChatGPT 4o was 0.755. Machine learning algorithms use Random Forest, Support Vector Machine, Logistic Regression, XGBoost and Decision Tree, the AUC of 5 machine learning algorithms was range from 0.534 to 0.624. ChatGPT 4o achieved the highest AUC and showed statistically significant differences compared to Support Vector Machine, Logistic Regression and Decision Tree. Additionally, the predictive results of the ChatGPT 4o effectively stratified the postoperative overall survival (OS) and recurrence-free survival (RFS) of HCC patients. LLMs have demonstrated significant predictive capabilities for MVI in HCC and for risk stratification regarding postoperative OS and RFS. These advancements possess substantial potential to enhance preoperative management and make surgical planning.
原发性肝癌是全球第六大最常被诊断出的癌症，也是癌症相关死亡的第三大主要原因。肝细胞癌（HCC）是原发性肝癌最常见的类型，微血管侵犯（MVI）是影响HCC术后预后的一个重要风险因素。然而，术前准确预测MVI仍然是一项挑战。本研究旨在评估大语言模型（LLMs），特别是ChatGPT 4o，在预测HCC中MVI的应用，并将其性能与传统临床模型进行比较。在这项回顾性研究中，在两个中心选取了2018年6月至2018年12月期间接受根治性肝切除术的300例HCC患者。收集的临床数据包括年龄、性别、乙肝病毒感染、肝硬化、甲胎蛋白水平等。使用ChatGPT 4o处理患者的临床数据并预测MVI。随后，将ChatGPT 4o的预测结果与机器学习模型进行比较，绘制ROC曲线并计算AUC。结果显示，ChatGPT 4o的AUC为0.755。机器学习算法使用随机森林、支持向量机、逻辑回归、XGBoost和决策树，5种机器学习算法的AUC范围为0.534至0.624。ChatGPT 4o的AUC最高，与支持向量机、逻辑回归和决策树相比有统计学显著差异。此外，ChatGPT 4o的预测结果有效地对HCC患者的术后总生存期（OS）和无复发生存期（RFS）进行了分层。大语言模型在预测HCC中的MVI以及术后OS和RFS的风险分层方面已显示出显著的预测能力。这些进展在加强术前管理和制定手术计划方面具有巨大潜力。
