多模态计算数据分析预测转移性黑色素瘤系统治疗反应
该回顾性研究纳入120例接受免疫检查点抑制剂或靶向治疗的转移性黑色素瘤患者,提取影像组学特征和身体成分分析生物标志物,构建基于临床、影像组学、影像-临床组学及联合身体成分分析的病灶级和患者级模型。结果显示整合治疗信息后所有患者级模型AUC均提升,联合模型AUC达0.97,显著优于临床模型(AUC 0.657)。
要点
- 整合治疗信息可提升所有患者级预测模型的AUC。
- 影像-临床组学联合身体成分分析模型AUC达0.97。
- 病灶特异性影像组学模型性能因部位而异,肝病灶AUC 0.98。
核心结论
- 整合多模态数据尤其是治疗信息和自动化身体成分分析特征,可显著提高转移性黑色素瘤治疗反应预测的准确性。
研究设计与样本
- 回顾性研究纳入120例接受免疫检查点抑制剂或靶向治疗的转移性黑色素瘤患者。
- 纳入标准包括治疗前后CT扫描中至少一个RECIST 1.1可识别病灶及包含治疗信息和临床特征的元数据。
- 提取影像组学特征和身体成分分析生物标志物,构建病灶级和患者级模型,分别在有和无治疗信息(类型和线数)条件下训练。
主要发现
- 所有患者级模型在纳入治疗信息后AUC均有改善。
- 临床模型AUC为0.657,影像组学模型AUC为0.92,影像-临床组学联合身体成分分析模型AUC为0.97。
- 病灶特异性影像组学模型性能不一,腹膜/软组织病灶AUC为0.76,肝病灶AUC为0.98。
临床怎么用
- 基于多模态数据(尤其是治疗信息和自动化身体成分分析特征)的预测模型有助于改善患者分层和个性化肿瘤诊疗。
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)。
原文
INTRODUCTION: Novel therapies like immunotherapy (ICI) and targeted therapy (TT) have significantly improved survival in metastatic melanoma, but treatment failure and/or development of treatment resistance pose major challenges. Radiomics has been shown to be a valuable tool for imaging-based and/or metadata-based (clinomics), non-invasive data analysis to predict therapy response in oncologic patients. This study aims to evaluate the predictive performance of clinical, radiomics, radio-clinomics, and combined radio-clinomics + body composition analysis (BCA) features for treatment response in metastatic melanoma patients undergoing ICI or TT. METHODS: A retrospective study was conducted on 120 metastatic melanoma patients receiving ICI or TT. Inclusion criteria comprised a minimum of one RECIST 1.1.-identified lesion in pre- and posttherapeutic CT scans as well as metadata including therapy information and clinical features. Radiomics features and BCA biomarkers were extracted. Lesion- and patient-specific models based on clinical, radiomics, radio-clinomics, and BCA parameters were trained twice: with and without therapy information (type and line). RESULTS: All patient-based models showed improved AUCs after inclusion of therapy information. The clinical model achieved modest AUC of 0.657, when compared to imaging included models with AUCs ranging from 0.92 (radiomics) to 0.97 (radio-clinomics + BCA). The lesion-specific radiomics model showed variable performance spanning from poor AUC of 0.76 in peritoneum/soft-tissue lesions to excellent AUC in liver lesions (AUC 0.98). CONCLUSION: Our results demonstrate that integrating multi-modal data, particularly therapy information and automated BCA features, substantially improves the accuracy of treatment response prediction, highlighting the potential for improved patient stratification and personalized oncologic care.
引用
- Parmar V, Lodde G, Haubold J, et al. Multi-modal computational data analysis for prediction of response to systemic treatment in metastatic melanoma. Eur J Cancer. 2026. PMID: 41895000. doi:10.1016/j.ejca.2026.116690
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。