利用深度迁移学习挖掘多参数Gd-EOB-DTPA增强磁共振成像特征构建HCC切除术后复发预测模型

HCC切除术后复发是导致患者治疗失败和死亡的主要原因，目前主要根据分期系统进行危险分层，但尚无法准确预测且不能提供可量化的风险指标。多参数Gd-EOB-DTPA增强MRI是临床诊疗HCC的主要依据，现有证据表明其数字特征对HCC切除术后复发预测很有价值。深度迁移学习的自学习特性能够自动挖掘图像更抽象、本质的特征，且显著减少用于训练的图像数量，但目前尚未有将深度迁移学习应用于多参数Gd-EOB-DTPA增强MRI图像挖掘特征以预测HCC复发风险的报道。本研究拟利用深度迁移学习提取多参数Gd-EOB-DTPA增强MRI图像数字特征,构建HCC自动分割模型，探索肿瘤、瘤周和背景肝实质图像特征与术后早、晚期复发之间的相关性;并结合MRI图像数字特征、分子标志物和临床-病理特征建立综合预测模型。预期可为临床提供更精确的HCC术后复发预测方法,为HCC患者个体化治疗方案制定提供决策支持。

HCC recurrence after resection is the main cause of treatment failure and death of patients. At present, risk stratification is mainly based on the HCC staging systems, however, they are criticized for inadequate predictive accuracy, and none of them provide quantifiable risk measures. Multi-parameter Gd-EOB-DTPA enhanced MRI is the main basis for clinical diagnosis and treatment of HCC. Existing evidence indicates that its digital characteristics are valuable for predicting recurrence after HCC resection. The self-learning feature of deep transfer learning can automatically mine the more abstract and essential features of images, and significantly reduce the number of images used for training. However, there has been no report that applies deep transfer learning to multi-parameter Gd-EOB-DTPA enhancement MRI for mining features and predicting the risk of HCC recurrence. Therefore，this study aims to use deep transfer learning to extract the digital features of multi-parameter Gd-EOB-DTPA enhanced MRI images for HCC patients, build an automatic segmentation model of HCC tumor areas, and explore the relationship between the features of tumor, peritumor and background liver parenchymal and postoperative early and late recurrence. Meanwhile, a comprehensive prediction model will be established by combining the digital characteristics of MRI images, related molecular markers and clinical-pathological characteristics. It is expected to provide a more accurate method for predicting postoperative recurrence of HCC in the clinic and provide decision support for the development of individualized treatment plans for HCC patients.