Abstract：
OBJECTIVE: This study explored the intratumor heterogeneity (ITH) of esophageal squamous cell carcinoma (ESCC) using computed tomography (CT) and investigated the value of CT-based ITH in predicting the response to immune checkpoint inhibitor (ICI) plus chemotherapy in patients with ESCC.
METHODS: This retrospective study included 416 patients with ESCC who received ICI plus chemotherapy at two independent hospitals between January 2019 and July 2022. Multiparametric CT features were extracted from ESCC lesions and screened using hierarchical clustering and dimensionality reduction algorithms. Logistic regression and machine learning models based on selected features were developed to predict treatment response and validated in separate datasets. ITH was quantified using the score calculated by the best-performing model and visualized through feature clustering and feature contribution heatmaps. A gene set enrichment analysis (GSEA) was performed to identify the biological pathways underlying the CT-based ITH.
RESULTS: The extreme gradient boosting model based on CT-derived ITH had higher discriminative power, with areas under the receiver operating characteristic curve of 0.864 (95% confidence interval [CI]: 0.774-0.954) and 0.796 (95% CI: 0.698-0.893) in the internal and external validation sets. The CT-based ITH pattern differed significantly between responding and non-responding patients. The GSEA indicated that CT-based ITH was associated with immunity-, keratinization-, and epidermal cell differentiation-related pathways.
CONCLUSIONS: CT-based ITH is an effective biomarker for identifying patients with ESCC who could benefit from ICI plus chemotherapy. Immunity-, keratinization-, and epidermal cell differentiation-related pathways may influence the patient\'s response to ICI plus chemotherapy.
METHODS: This retrospective study included 416 patients with ESCC who received ICI plus chemotherapy at two independent hospitals between January 2019 and July 2022. Multiparametric CT features were extracted from ESCC lesions and screened using hierarchical clustering and dimensionality reduction algorithms. Logistic regression and machine learning models based on selected features were developed to predict treatment response and validated in separate datasets. ITH was quantified using the score calculated by the best-performing model and visualized through feature clustering and feature contribution heatmaps. A gene set enrichment analysis (GSEA) was performed to identify the biological pathways underlying the CT-based ITH.
RESULTS: The extreme gradient boosting model based on CT-derived ITH had higher discriminative power, with areas under the receiver operating characteristic curve of 0.864 (95% confidence interval [CI]: 0.774-0.954) and 0.796 (95% CI: 0.698-0.893) in the internal and external validation sets. The CT-based ITH pattern differed significantly between responding and non-responding patients. The GSEA indicated that CT-based ITH was associated with immunity-, keratinization-, and epidermal cell differentiation-related pathways.
CONCLUSIONS: CT-based ITH is an effective biomarker for identifying patients with ESCC who could benefit from ICI plus chemotherapy. Immunity-, keratinization-, and epidermal cell differentiation-related pathways may influence the patient\'s response to ICI plus chemotherapy.
摘要：
目的：本研究利用计算机断层扫描（CT）技术探讨食管鳞状细胞癌（ESCC）的瘤内异质性（ITH），并探讨基于CT的ITH在预测ESCC患者对免疫检查点抑制剂（ICI）加化疗反应中的价值。
方法：这项回顾性研究纳入了在2019年1月至2022年7月期间在两家独立医院接受ICI加化疗的416例ESCC患者。从ESCC病变中提取多参数CT特征，并使用层次聚类和降维算法进行筛选。开发了基于选定特征的逻辑回归和机器学习模型来预测治疗反应，并在单独的数据集中进行了验证。ITH使用由表现最好的模型计算的分数进行量化,并通过特征聚类和特征贡献热图进行可视化。进行基因集富集分析(GSEA)以鉴定基于CT的ITH的生物途径。
结果：基于CT推导的ITH的极值梯度提升模型具有更高的判别能力，在内部和外部验证集中,受试者工作特征曲线下面积分别为0.864(95%置信区间[CI]:0.774-0.954)和0.796(95%CI:0.698-0.893).基于CT的ITH模式在有反应和无反应的患者之间存在显着差异。GSEA表明基于CT的ITH与免疫力相关，角化-，和表皮细胞分化相关途径。
结论：基于CT的ITH是一种有效的生物标志物，可用于识别可能从ICI加化疗中受益的ESCC患者。免疫力-，角化-，与表皮细胞分化相关的通路可能会影响患者对ICI+化疗的反应。
方法：这项回顾性研究纳入了在2019年1月至2022年7月期间在两家独立医院接受ICI加化疗的416例ESCC患者。从ESCC病变中提取多参数CT特征，并使用层次聚类和降维算法进行筛选。开发了基于选定特征的逻辑回归和机器学习模型来预测治疗反应，并在单独的数据集中进行了验证。ITH使用由表现最好的模型计算的分数进行量化,并通过特征聚类和特征贡献热图进行可视化。进行基因集富集分析(GSEA)以鉴定基于CT的ITH的生物途径。
结果：基于CT推导的ITH的极值梯度提升模型具有更高的判别能力，在内部和外部验证集中,受试者工作特征曲线下面积分别为0.864(95%置信区间[CI]:0.774-0.954)和0.796(95%CI:0.698-0.893).基于CT的ITH模式在有反应和无反应的患者之间存在显着差异。GSEA表明基于CT的ITH与免疫力相关，角化-，和表皮细胞分化相关途径。
结论：基于CT的ITH是一种有效的生物标志物，可用于识别可能从ICI加化疗中受益的ESCC患者。免疫力-，角化-，与表皮细胞分化相关的通路可能会影响患者对ICI+化疗的反应。
