文献速递腹部技术与安全CT

基于对比增强CT与组织病理学的多模态深度学习模型预测结直肠癌微卫星不稳定性

Multimodal deep learning model for predicting microsatellite instability in colorectal cancer by contrast-enhanced computed tomography and histopathology.
PubMed 文献 · Eur J Radiol · 2025-10-12

本研究为回顾性多中心研究,纳入305例有配对对比增强CT和全切片图像的结直肠癌患者,构建融合静脉期CT与病理特征的多模态深度学习模型预测微卫星不稳定状态。融合模型在外部测试集AUC为0.905,优于单模态及其他融合模型,结论认为该模型泛化性稳健,可支持患者分层和免疫治疗决策。

要点

  • 融合静脉期CT与病理深度学习的模型在外部测试集预测MSI的AUC为0.905。
  • 外部测试集准确率84.2%、敏感度80.3%、特异度83.7%、F1分数0.837。
  • 融合模型外部测试集AUC优于病理模型0.794、静脉期CT模型0.858、APDL 0.802和F-AVPDL 0.813。

核心结论

  • F-VP-PathDL模型跨中心泛化性稳健,为结直肠癌MSI预测提供可临床扩展的工具,支持患者分层并辅助免疫治疗决策。

研究设计与样本

  • 回顾性多中心研究,纳入305例有配对术前对比增强CT和术后全切片图像的结直肠癌患者。
  • 中心I和II患者分配至训练集169例和内部验证集85例,中心III患者构成外部测试集51例。
  • 病理深度学习模型(PathDL)采用EfficientNet-b0架构,静脉期CT深度学习模型(VPDL)采用ResNet 101架构。
  • 融合模型F-VP-PathDL采用自适应残差网络整合两种模态特征。
  • 模型性能以受试者工作特征曲线下面积(AUC)、准确率、敏感度、特异度和F1分数评估。

诊断标准与关键数值

  • 内部验证集F-VP-PathDL模型AUC为0.883(95%CI:0.732-0.967)。
  • 外部测试集F-VP-PathDL模型AUC为0.905(95%CI:0.831-0.945)。
  • 外部测试集准确率84.2%(95%CI:69.1%-92.8%),敏感度80.3%(95%CI:28.4%-98.7%),特异度83.7%(95%CI:68.8%-93.9%),F1分数0.837(95%CI:0.326-0.999)。
  • 外部测试集AUC比较:F-VP-PathDL 0.905 vs PathDL 0.794、VPDL 0.858、APDL 0.802、F-AVPDL 0.813。

临床应用与阅片要点

  • 原文结论认为该模型可作为结直肠癌MSI预测的可临床扩展工具,支持患者分层并辅助免疫治疗决策。

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: To develop and validate a multimodal deep learning (DL) model that integrates preoperative contrast-enhanced computed tomography (CECT) and postoperative whole-slide images (WSIs) to predict microsatellite instability (MSI) status in colorectal cancer (CRC).

MATERIALS AND METHODS: This retrospective, multicenter study enrolled 305 CRC patients with paired CECT and WSIs. Patients from Center I and II were allocated to the training (n = 169) and internal validation (n = 85) sets, while those from Center III formed the external test set (n = 51). Pathology-based DL (PathDL) and venous-phase CECT (VPDL) models were constructed using EfficientNet-b0 and ResNet 101 architectures, respectively. A fusion model (F-VP-PathDL, Fusion of venous phase CT and pathology with deep learning) was developed using an adaptive residual network to integrate features from both modalities. Model performance was evaluated using area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, and F1 score.

RESULTS: The F-VP-PathDL model achieved strong performance on the internal validation set, with an AUC of 0.883 (95 % CI: 0.732-0.967). On the external test set, the model achieved an AUC of 0.905 (95 % CI: 0.831-0.945), outperforming single-modality and alternative fusion models (PathDL: 0.794; VPDL: 0.858; APDL: 0.802; F-AVPDL: 0.813). The model also demonstrated robust accuracy (84.2 %, 95 % CI: 69.1 %-92.8 %), sensitivity (80.3 %, 95 % CI: 28.4 %-98.7 %), specificity (83.7 %, 95% CI: 68.8 %-93.9 %) and F1 score (0.837, 95 % CI: 0.326-0.999) on the external test set.

CONCLUSIONS: The F-VP-PathDL model demonstrates robust generalizability across centers and offers a clinically scalable tool for MSI prediction in CRC, supporting patient stratification and informing immunotherapy decisions.

来源:Eur J Radiol,查看原文。版权归原作者所有。

引用

  1. Tang K, She R, Chen G, et al. Multimodal deep learning model for predicting microsatellite instability in colorectal cancer by contrast-enhanced computed tomography and histopathology. Eur J Radiol. 2025. PMID: 41101004. doi:10.1016/j.ejrad.2025.112468

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。