Department of Radiology, Guangdong Provincial People's Hospital, Guangdong Academy of Medical Sciences, Guangzhou, China.
School of Medical Information Engineering, Guangzhou University of Chinese Medicine, Guangzhou, China.
Acad Radiol. 2021 Nov;28 Suppl 1:S118-S127. doi: 10.1016/j.acra.2020.11.014. Epub 2020 Dec 7.
To investigate the value of diffusion-weighted magnetic resonance imaging for the prediction of microvascular invasion (MVI) of Hepatocellular Carcinoma (HCC) using Convolutional Neural Networks (CNN).
This study was approved by the local institutional review board and the patients' informed consent was waived. Consecutive 97 subjects with 100 HCCs from July 2012 to October 2018 with surgical resection were retrieved. All subjects with diffusion-weighted imaging (DWI) examinations were performed with single-shot echo-planar imaging in a breath-hold routine. DWI parameters were three b values of 0,100,600 sec/mm. First, apparent diffusion coefficients (ADC) images were computed by mono-exponentially fitting the three b-value points. Then, multiple 2D axial patches (28 × 28) of HCCs from b0, b100, b600, and ADC images were extracted to increase the dataset for training the CNN model. Finally, the fusion of deep features derived from three b value images and ADC was conducted based on the CNN model for MVI prediction. The data set was split into the training set (60 HCCs) and the independent test set (40 HCCs). The output probability of the deep learning model in the MVI prediction of HCCs was assessed by the independent student's t-test for data following a normal distribution and Mann-Whitney U test for data violating the normal distribution. Receiver operating characteristic curve and area under the curve (AUC) were also used to assess the performance for MVI prediction of HCCs in the fixed test set.
Deep features in b600 images yielded better performance (AUC = 0.74, p = 0.004) for MVI prediction than b0 (AUC = 0.69, p = 0.023) and b100 (AUC = 0.734, p = 0.011). Comparatively, deep features in the ADC map obtained lower performance (AUC = 0.71, p = 0.012) than that of the higher b value images (b600) for MVI prediction. Furthermore, the fusion of deep features from the b0, b100, b600, and ADC images yielded the best results (AUC = 0.79, p = 0.002) for MVI prediction.
Fusion of deep features derived from DWI images concerning the three b-value images and the ADC image yields better performance for MVI prediction.
利用卷积神经网络（CNN）探讨弥散加权磁共振成像（DWI）对肝细胞癌（HCC）微血管侵犯（MVI）的预测价值。
本研究经当地机构审查委员会批准，豁免了患者的知情同意。回顾性分析 2012 年 7 月至 2018 年 10 月期间行手术切除的 97 例 HCC 患者（共 100 个病灶）的临床资料。所有患者均行单次激发回波平面弥散加权成像（DWI）检查，屏气状态下完成。DWI 采用三个弥散敏感系数（b 值）：0、100、600 sec/mm2。首先，通过对三个 b 值点进行单指数拟合得到表观弥散系数（ADC）图像。然后，从 b0、b100、b600 和 ADC 图像中提取 HCC 的多个 2D 轴向斑块（28×28），以增加训练 CNN 模型的数据量。最后，基于 CNN 模型融合来自三个 b 值图像和 ADC 的深度特征以进行 MVI 预测。数据集分为训练集（60 个 HCC）和独立测试集（40 个 HCC）。通过独立样本 t 检验（数据符合正态分布时）和 Mann-Whitney U 检验（数据不符合正态分布时）评估深度学习模型在 HCCs 的 MVI 预测中的输出概率。在固定测试集中，还使用受试者工作特征曲线和曲线下面积（AUC）评估 HCCs 的 MVI 预测性能。
与 b0（AUC=0.69，p=0.023）和 b100（AUC=0.734，p=0.011）相比，b600 图像的深度特征在 MVI 预测中具有更好的性能（AUC=0.74，p=0.004）。相比之下，ADC 图中的深度特征在 MVI 预测中的性能较低（AUC=0.71，p=0.012）。此外，来自 b0、b100、b600 和 ADC 图像的深度特征融合在 MVI 预测中产生了最佳结果（AUC=0.79，p=0.002）。
DWI 图像的三个 b 值图像和 ADC 图像的深度特征融合对 MVI 预测具有更好的性能。
