Radiomic-Based Pathological Response Prediction from Primary Tumors and Lymph Nodes in NSCLC.
Radiomic-Based Pathological Response Prediction from Primary Tumors and Lymph Nodes in NSCLC.
复制标题
DOI：
10.1016/j.jtho.2016.11.2226
复制
发表时间：
2017-03
期刊：
影响因子：
--
通讯作者：
Aerts HJWL
中科院分区：
文献类型：
--
作者：
Coroller TP;Agrawal V;Huynh E;Narayan V;Lee SW;Mak RH;Aerts HJWL
捕获总肿瘤负荷的非侵入性生物标志物可以为精准医学提供重要的补充信息，以帮助临床决策。我们研究了从原发肿瘤和淋巴结的治疗前计算机断层扫描（CT）图像中提取的放射组学数据在预测手术前新辅助放化疗后的病理反应中的价值。本IRB批准的研究纳入了2003-2013年接受治疗的85例可切除局部晚期（II-III期）非小细胞肺癌（NSCLC）患者（中位年龄：60.3岁; 65%为女性）。对85个原发性肿瘤和178个淋巴结进行放射组学分析，以区分病理完全缓解（pCR）或大体残留疾病（GRD）。使用曲线下面积（AUC）评价了20个非冗余和稳定特征（每个研究中心10个）（针对多重假设检验校正了所有p值）。使用随机森林和嵌套交叉验证评估每个特征集的分类性能。三个放射组学特征（描述原发性肿瘤球形度和淋巴结同质性）显著预测具有相似性能的pCR（所有AUC=0.67，p值<0.05）。这两个特征（量化淋巴结同质性）预测GRD（AUC范围：0.72-0.75，p值<0.05），并且表现显著优于主要特征（AUC=0.62）。多变量分析显示，对于pCR，单独的放射组学特征集具有最佳的分类性能（中位AUC=0.68）。此外，对于GRD分类，结合放射组学和临床数据显著优于所有其他特征集（中位AUC=0.73）。淋巴结表型信息是显着预测病理反应，并表现出更高的分类性能比放射组学特征获得的原发性肿瘤。
Non-invasive biomarkers that capture the total tumor burden could provide important complementary information for precision medicine to aid clinical decision-making. We investigated the value of radiomic data, extracted from pre-treatment computed tomography (CT) images of the primary tumor and lymph nodes, in predicting pathological response following neoadjuvant chemoradiation prior to surgery. Eighty-five patients with resectable locally-advanced (stage II-III) non-small cell lung cancer (NSCLC) (median age: 60.3 years; 65% female) treated from 2003-2013 were included in this IRB-approved study. Radiomics analysis was performed on 85 primary tumors and 178 lymph nodes to discriminate between pathological complete response (pCR) or gross residual disease (GRD). Twenty non-redundant and stable features (10 from each site) were evaluated using the area under the curve (AUC) (all p-values were corrected for multiple hypothesis testing). Classification performance of each feature set was evaluated using random forest and nested cross validation. Three radiomic features (describing primary tumor sphericity and lymph node homogeneity) were significantly predictive of pCR with similar performances (all AUC=0.67, p-value<0.05). Those two features (quantifying lymph node homogeneity) were predictive of GRD (AUC range: 0.72-0.75, p-value<0.05) and performed significantly better than the primary features (AUC=0.62). Multivariate analysis showed that for pCR, the radiomics feature set alone had the best performing classification (median AUC=0.68). Furthermore, for GRD classification, combining radiomic and clinical data significantly outperformed all other feature sets (median AUC=0.73). Lymph node phenotypic information was significantly predictive for pathological response and showed higher classification performance than radiomic features obtained from the primary tumor.