Department of Radiology, CHU de Bordeaux, F-33000, France.
University of Bordeaux, IMB, UMR CNRS 5251, INRIA Project team Monc, 33400 Talence, France.
Diagn Interv Imaging. 2022 Jul-Aug;103(7-8):360-366. doi: 10.1016/j.diii.2022.01.009. Epub 2022 Feb 17.
The purpose of this study was to evaluate the capabilities of radiomics using magnetic resonance imaging (MRI) data in the assessment of treatment response to yttrium transarterial radioembolization (TARE) in patients with locally advanced hepatocellular carcinoma (HCC) by comparison with predictions based on European Association for the Study of the Liver (EASL) criteria.
Twenty-two patients with HCC (19 men, 3 women; mean age: 66.7 ± 9.8 [SD]; age range: 37-82 years) who underwent contrast-enhanced MRI 4 ± 1 weeks before and 4 ± 4 weeks after TARE, were enrolled in this retrospective study. Regions of interest were placed manually along the contours of the treated tumor on each axial slice of arterial and portal phase images using the ITK-SNAP post-processing software. For each MRI, the Pyradiomics Python package was used to extract 107 radiomics features on both arterial and portal phases, and resulting delta-features were computed. The Mann-Whitney U test with Bonferroni correction was used to select statistically different features between responders and non-responders (i.e., those with progressive or stable disease) at 6-month follow-up, according to the modified Response Evaluation Criteria in Solid Tumors (mRECIST). Finally, for each selected feature, univariable logistic regression with leave-one-out cross validation procedure was used to perform receiver operating characteristic (ROC) curve analysis and compare radiomics parameters with MRI variables.
According to mRECIST, 14 patients (14/22; 64%) were non-responders and 8 (8/22; 36%) were responders. Four radiomics parameters (long run emphasis, minor axis length, surface area, and gray level non-uniformity on arterial phase images) were the only predictors of early response. ROC curve analysis showed that long run emphasis was the best parameter for predicting early response, with 100% sensitivity (95% CI: 68-100) and 100% specificity (95% CI: 78-100). EASL morphologic criteria yielded 75% sensitivity (95% CI: 41-96%) and 93% specificity (95% CI: 69-100%).
Radiomics allows identify marked differences between responders and non-responders, and could aid in the prediction of early treatment response following TARE in patients with HCC.
本研究旨在通过与基于欧洲肝脏研究协会（EASL）标准的预测结果进行比较，评估磁共振成像（MRI）数据的放射组学在评估局部晚期肝细胞癌（HCC）患者钇经动脉放射性栓塞（TARE）治疗反应中的能力。
本回顾性研究纳入了 22 例 HCC 患者（19 名男性，3 名女性；平均年龄 66.7±9.8[标准差]；年龄范围 37-82 岁），他们在 TARE 前 4±1 周和 TARE 后 4±4 周进行了对比增强 MRI 检查。使用 ITK-SNAP 后处理软件，在动脉期和门静脉期的每个轴位切片上手动勾画感兴趣区（ROI）。对于每个 MRI，使用 Pyradiomics Python 包在动脉期和门静脉期提取 107 个放射组学特征，并计算出相应的 delta 特征。采用曼-惠特尼 U 检验（Mann-Whitney U test）和 Bonferroni 校正，根据改良实体瘤反应评价标准（mRECIST），在 6 个月随访时，选择具有统计学差异的特征，将其用于评估治疗反应。最后，对于每个选定的特征，使用单变量逻辑回归（univariable logistic regression）和留一法交叉验证程序进行受试者工作特征（ROC）曲线分析，并比较放射组学参数与 MRI 变量。
根据 mRECIST，14 例患者（14/22；64%）为无应答者，8 例患者（8/22；36%）为应答者。4 个放射组学参数（动脉期图像的长行程强调、短轴长度、表面积和灰度不均匀性）是早期应答的唯一预测因素。ROC 曲线分析显示，长行程强调是预测早期应答的最佳参数，其敏感性为 100%（95%可信区间：68-100%），特异性为 100%（95%可信区间：78-100%）。EASL 形态学标准的敏感性为 75%（95%可信区间：41-96%），特异性为 93%（95%可信区间：69-100%）。
放射组学可以识别出应答者和无应答者之间的显著差异，并有助于预测 HCC 患者 TARE 后的早期治疗反应。
