基于深度学习影像组学列线图自动预测晚期肝细胞癌肝动脉灌注化疗反应
本研究回顾性纳入三家医院458例接受FOLFOX方案肝动脉灌注化疗的晚期肝细胞癌患者,基于增强CT自动分割肝区提取深度学习和影像组学特征构建DLRN模型。DLRN在训练组、内部验证组和外部验证组的AUC分别为0.988、0.915和0.896,预测客观缓解组的总体生存期显著长于预测非缓解组(中位OS 26.0 vs 12.3个月,p<0.001)。
要点
- DLRN基于增强CT自动分割肝区,整合深度学习评分、影像组学评分及临床变量预测HAIC客观缓解。
- DLRN在训练组、内部验证组和外部验证组的AUC分别为0.988、0.915和0.896。
- 预测客观缓解组的中位OS为26.0个月,显著长于预测非缓解组的12.3个月(p<0.001)。
核心结论
- DLRN对预测晚期肝细胞癌HAIC反应表现出满意性能,有助于筛选适合HAIC的患者,并可能使个体化治疗前决策获益。
研究设计与样本
- 回顾性纳入三家医院458例接受FOLFOX方案HAIC的晚期肝细胞癌患者,310例为训练组、77例为内部验证组、71例为外部验证组;从增强CT自动分割肝区提取深度学习和影像组学特征,以多因素logistic回归整合深度学习评分、影像组学评分及显著临床变量构建DLRN;以AUC和Kaplan-Meier估计评估模型性能。
诊断标准与关键数值
- DLRN预测HAIC客观缓解的AUC在训练组为0.988、内部验证组为0.915、外部验证组为0.896。
- DLRN预测的客观缓解组中位OS为26.0个月,预测的非客观缓解组为12.3个月,p<0.001。
临床应用与阅片要点
- DLRN可基于增强CT自动分割肝区进行HAIC反应预测,并用于生存风险分层。
- DLRN为晚期肝细胞癌患者治疗前个体化决策的易用工具。
局限性
- 原文未报告局限性。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
英文原文(摘要)· 点击展开
OBJECTIVES: Hepatic arterial infusion chemotherapy (HAIC) using the FOLFOX regimen (oxaliplatin plus fluorouracil and leucovorin) is a promising option for advanced hepatocellular carcinoma (Ad-HCC). As identifying patients with Ad-HCC who would obtain objective response (OR) to HAIC preoperatively remains a challenge, we aimed to develop an automatic and non-invasive model for predicting HAIC response.
METHODS: A total of 458 patients with Ad-HCC who underwent HAIC were retrospectively included from three hospitals (310 for training, 77 for internal validation, and 71 for external validation). The deep learning and radiomic features were extracted from the automatically segmented liver region on contrast-enhanced computed tomography images. Then, a deep learning radiomic nomogram (DLRN) was constructed by integrating deep learning scores, radiomic scores, and significant clinical variables with multivariate logistic regression. Model performance was assessed by AUC and Kaplan-Meier estimator.
RESULTS: After automatic segmentation, only a few modifications were needed (less than 30 min for 458 patients). The DLRN achieved an AUC of 0.988 in the training cohort, 0.915 in the internal validation cohort, and 0.896 in the external validation cohort, respectively, outperforming other models in HAIC response prediction. Moreover, survival risk stratification was also successfully performed by the DLRN. The overall survival (OS) of the predictive OR group was significantly longer than that of the predictive non-OR group (median OS: 26.0 vs. 12.3 months, p < 0.001).
CONCLUSIONS: The DLRN provided a satisfactory performance for predicting HAIC response, which is essential to identify Ad-HCC patients for HAIC and may potentially benefit personalized pre-treatment decision-making.
CLINICAL RELEVANCE STATEMENT: This study presents an accurate and automatic method for predicting response to hepatic arterial infusion chemotherapy in patients with advanced hepatocellular carcinoma, and therefore help in defining the best candidates for this treatment.
KEY POINTS: • Deep learning radiomic nomogram (DLRN) based on automatic segmentation of CECT can accurately predict hepatic arterial infusion chemotherapy (HAIC) response of advanced HCC patients. • The proposed prediction model can perform survival risk stratification and is an easy-to-use tool for personalized pre-treatment decision-making for advanced HCC patients.
引用
- Xu Z, An C, Shi F, et al. Automatic prediction of hepatic arterial infusion chemotherapy response in advanced hepatocellular carcinoma with deep learning radiomic nomogram. Eur Radiol. 2023. PMID: 37498380. doi:10.1007/s00330-023-09953-x
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。