胰腺导管腺癌肝转移的CT纹理分析预测模型
本研究回顾性纳入220例经病理证实的初治胰腺导管腺癌患者,基于原发灶CT动脉期和门脉期纹理特征构建肝转移预测模型。模型含8个变量,逻辑回归模型有统计学意义(χ2=81.6,p<.001),正确分类80.9%病例,敏感度58.6%、特异度91.3%、AUC 0.850(95%CI 0.793-0.907)。肿瘤大小、动脉期HU_MAX、动脉期GLZLM_SZHGE和门脉期GLCM_CORRELATION为显著预测因子。
要点
- CT纹理分析可辅助预测PDAC肝转移风险,模型特异度较高(91.3%)。
- 肿瘤大小、动脉期HU_MAX、动脉期GLZLM_SZHGE、门脉期GLCM_CORRELATION为独立预测因子。
- 模型敏感度中等(58.6%),阴性预测值82.5%,需结合临床综合判断。
关键影像表现/征象
- 肿瘤大小、动脉期HU_MAX、动脉期GLRLM_LRLGE、动脉期GLZLM_SZHGE、动脉期GLZLM_LZLGE、门脉期GLCM_CORRELATION、门脉期GLRLM_LRLGE和门脉期GLZLM_SZHGE在转移组与非转移组间有显著差异。
诊断标准与关键数值
- 逻辑回归模型χ2=81.6,p<.001,正确分类80.9%病例。
- 模型敏感度58.6%、特异度91.3%、阳性预测值75.9%、阴性预测值82.5%。
- AUC为0.850(95%CI 0.793-0.907)。
- 肿瘤大小、动脉期HU_MAX、动脉期GLZLM_SZHGE、门脉期GLCM_CORRELATION的OR分别为1.1、0.9、1、1.49。
临床应用与阅片要点
- 阅片时可关注原发灶CT纹理特征,辅助评估肝转移风险。
- 模型特异度高,阳性结果对肝转移的提示价值较大,但敏感度有限。
局限性
- 回顾性研究,可能存在选择偏倚。
- 单中心研究,模型需外部验证。
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
PURPOSE: To develop a predictive model for liver metastases in patients with pancreatic ductal adenocarcinoma (PDAC) based on textural features of the primary tumor extracted by computed tomography (CT) images.
MATERIALS AND METHODS: Patients with a pathologically proved PDAC who underwent CT between December 2020 and January 2022 were retrospectively identified. Treatment-naïve patients were included. Sex, age, tumor size, vascular infiltration and 39 arterial and portal phase textural features were analyzed. The variables significantly correlated to tumor size according to the Pearson's product-moment correlation test were excluded from analysis; the remaining variables were compared between metastatic (M +) and non-metastatic (M-) patients using Fisher's or Mann-Whitney test. The features with a significant difference between groups were entered into a binomial logistic regression test to develop a predictive model for liver metastases.
RESULTS: This study included 220 patients. Eight variables (tumor size, arterial HU_MAX, arterial GLRLM_LRLGE, arterial GLZLM_SZHGE, arterial GLZLM_LZLGE, portal GLCM_CORRELATION, portal GLRLM_LRLGE, and portal GLZLM_SZHGE) were significantly different between groups. The logistic regression model was statistically significant (χ2 = 81.6, p < .001) and correctly classified 80.9% of cases. Sensitivity, specificity, positive and negative predictive values of the model were 58.6%, 91.3%, 75.9% and 82.5%, respectively. The area under the ROC curve of the model was 0.850 (95% CI, 0.793-0.907). Tumor size, arterial HU_MAX, arterial GLZLM_SZHGE and portal GLCM_CORRELATION were significant predictors of the likelihood of liver metastases, with odds ratios of 1.1, 0.9, 1, and 1.49, respectively.
CONCLUSIONS: CT texture analysis of PDAC can identify features that may predict the likelihood of liver metastases.
引用
- De Robertis, Geraci, Tomaiuolo, et al. Liver metastases in pancreatic ductal adenocarcinoma: a predictive model based on CT texture analysis. Radiol Med. 2022. PMID: 36057929. doi:10.1007/s11547-022-01548-8
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。