Dataset related to article "PET/CT-based radiomics of mass-forming intrahepatic cholangiocarcinoma improves prediction of pathology data and survival "

This record contains raw data related to article "PET/CT-based radiomics of mass-forming intrahepatic cholangiocarcinoma improves prediction of pathology data and survival"

Abstract

Purpose: Intrahepatic cholangiocarcinoma (IHC) is an aggressive disease with few reliable preoperative biomarkers. This study aims to elucidate if radiomics extracted from preoperative [18F]FDG PET/CT may grant a non-invasive biological characterization of IHC and predict outcome after complete resection of the tumor.

Methods: All patients preoperatively imaged by [18F]FDG PET/CT who underwent hepatectomy for mass-forming IHC in the period 2010-2019 were retrospectively evaluated. On PET images, manual slice-by-slice segmentation of IHC was performed (Tumor-VOI). A 5-mm margin region was semi-automatically generated around the tumor (Margin-VOI). Textural analysis was performed using the LifeX software. Analyzed outcomes included tumor grading (G3 vs. G1-2), microvascular invasion (MVI), overall survival (OS), and progression-free survival (PFS). The performances of the combined clinical-radiomic models were compared with those of standard clinical models.

Results: Overall, 74 patients (40 females, median age 68 years) were included. Considering tumor grading and MVI, the models combining the clinical data and radiomics of the Tumor-VOI had better performances than the clinical ones (AUC = 0.78 vs. 0.72 for grading; 0.87 vs. 0.78 for MVI). The inclusion into the models of radiomics of the Margin-VOI further improved the prediction of grading (AUC = 0.83), but not of MVI. Considering OS and PFS, the models including the preoperative clinical data and radiomics of the Tumor-VOI and Margin-VOI had better performances than the pure clinical ones (C-index = 0.81 vs. 0.76 for OS; 0.81 vs. 0.72 for PFS) and similar to the models including the pathology and postoperative data (C-index = 0.81 for OS; 0.79 for PFS). No model retained the standard SUV measures.

Conclusion: The PET-based radiomics of IHC can predict pathology data and allow a reliable preoperative evaluation of prognosis. The radiomics of both the tumoral and peritumoral areas had clinical relevance. The combined clinical-radiomic models outperformed the pure preoperative clinical ones and achieved performances non-inferior to the postoperative models.

本数据集关联论文《基于PET/CT的肿块型肝内胆管癌（intrahepatic cholangiocarcinoma, IHC）放射组学（radiomics）可提升病理数据与生存预后预测效能》的原始研究数据。

摘要

目的：肝内胆管癌（intrahepatic cholangiocarcinoma, IHC）是一种侵袭性极强的恶性疾病，目前临床可用的可靠术前生物标志物极为匮乏。本研究旨在探讨从术前[18F]FDG PET/CT影像中提取的放射组学特征，能否实现肝内胆管癌的无创生物学表征，并预测肿瘤完全切除术后的患者预后结局。

方法：本研究回顾性分析了2010至2019年间，因肿块型肝内胆管癌接受肝切除术、且术前接受[18F]FDG PET/CT影像学检查的全部患者。在PET影像上，研究者对肿瘤进行逐层手动分割，得到肿瘤体积感兴趣区（Tumor-VOI）；并围绕肿瘤半自动生成5mm宽度的瘤周边缘区域，即边缘感兴趣区（Margin-VOI）。采用LifeX软件完成纹理特征分析。本研究的分析结局指标包括肿瘤分级（G3级 vs G1-2级）、微血管侵犯（microvascular invasion, MVI）、总生存期（overall survival, OS）以及无进展生存期（progression-free survival, PFS）。将临床-放射组学联合模型的预测性能与纯临床模型进行对比。

结果：本研究共纳入74例患者（其中女性40例，中位年龄68岁）。针对肿瘤分级与微血管侵犯预测任务，结合临床数据与肿瘤体积感兴趣区放射组学特征的联合模型，其预测性能优于纯临床模型（分级预测的曲线下面积（area under the curve, AUC）为0.78 vs 0.72；微血管侵犯预测的AUC为0.87 vs 0.78）。若将边缘感兴趣区的放射组学特征纳入模型，则可进一步提升肿瘤分级的预测效能（AUC升至0.83），但对微血管侵犯的预测无显著增益。针对总生存期与无进展生存期预测任务，结合术前临床数据、肿瘤体积感兴趣区及边缘感兴趣区放射组学特征的联合模型，其性能优于纯术前临床模型（总生存期的一致性指数（C-index）为0.81 vs 0.76；无进展生存期的C-index为0.81 vs 0.72），且与纳入病理及术后数据的模型性能相当（总生存期C-index为0.81；无进展生存期C-index为0.79）。所有模型均未保留标准化摄取值（SUV）相关测量指标。

结论：基于PET影像的肝内胆管癌放射组学特征，可有效预测病理相关指标并实现可靠的术前预后评估。肿瘤区域与瘤周区域的放射组学特征均具备临床应用价值。临床-放射组学联合模型的预测性能优于纯术前临床模型，且其效能不逊于纳入术后及病理数据的模型。