Department of Hepatobiliary and Pancreatic Surgery, The First Hospital of Jilin University, Changchun, People's Republic of China.
College of Computer Science and Technology, Jilin University, Changchun, People's Republic of China.
Ann Surg Oncol. 2022 Oct;29(11):6774-6783. doi: 10.1245/s10434-022-12000-6. Epub 2022 Jun 26.
Hepatocellular carcinoma (HCC) is the fourth most common cause of cancer death worldwide, and the prognosis remains dismal. In this study, two pivotal factors, microvascular invasion (MVI) and vessels encapsulating tumor clusters (VETC) were preoperatively predicted simultaneously to assess prognosis.
A total of 133 HCC patients who underwent surgical resection and preoperative gadolinium ethoxybenzyl-diethylenetriaminepentaacetic acid (Gd-EOB-DTPA)-enhanced magnetic resonance imaging (MRI) were included. The statuses of MVI and VETC were obtained from the pathological report and CD34 immunohistochemistry, respectively. A three-dimensional convolutional neural network (3D CNN) for single-task learning aimed at MVI prediction and for multitask learning aimed at simultaneous prediction of MVI and VETC was established by using multiphase Gd-EOB-DTPA-enhanced MRI.
The 3D CNN for single-task learning achieved an area under receiver operating characteristics curve (AUC) of 0.896 (95% CI: 0.797-0.994). Multitask learning with simultaneous extraction of MVI and VETC features improved the performance of MVI prediction, with an AUC value of 0.917 (95% CI: 0.825-1.000), and achieved an AUC value of 0.860 (95% CI: 0.728-0.993) for the VETC prediction. The multitask learning framework could stratify high- and low-risk groups regarding overall survival (p < 0.0001) and recurrence-free survival (p < 0.0001), revealing that patients with MVI+/VETC+ were associated with poor prognosis.
A deep learning framework based on 3D CNN for multitask learning to predict MVI and VETC simultaneously could improve the performance of MVI prediction while assessing the VETC status. This combined prediction can stratify prognosis and enable individualized prognostication in HCC patients before curative resection.
肝细胞癌（HCC）是全球第四大常见癌症死因，预后仍然较差。在本研究中，同时预测了微血管侵犯（MVI）和血管包裹肿瘤簇（VETC）这两个关键因素，以评估预后。
共纳入 133 例接受手术切除和术前钆乙氧基苯甲基-二乙三胺五乙酸（Gd-EOB-DTPA）增强磁共振成像（MRI）检查的 HCC 患者。MVI 和 VETC 的状态分别从病理报告和 CD34 免疫组化获得。建立了一个用于单任务学习的三维卷积神经网络（3D CNN），旨在预测 MVI，以及一个用于同时预测 MVI 和 VETC 的多任务学习的 3D CNN，使用多期 Gd-EOB-DTPA 增强 MRI。
用于单任务学习的 3D CNN 的受试者工作特征曲线（ROC）下面积（AUC）为 0.896（95%可信区间：0.797-0.994）。同时提取 MVI 和 VETC 特征的多任务学习提高了 MVI 预测的性能，AUC 值为 0.917（95%可信区间：0.825-1.000），VETC 预测的 AUC 值为 0.860（95%可信区间：0.728-0.993）。多任务学习框架可以根据总体生存（p < 0.0001）和无复发生存（p < 0.0001）对高风险和低风险组进行分层，表明 MVI+/VETC+患者预后较差。
一种基于 3D CNN 的深度学习框架，用于同时预测 MVI 和 VETC，可以提高 MVI 预测的性能，同时评估 VETC 状态。这种联合预测可以对 HCC 患者在根治性切除前进行预后分层，实现个体化预测。
