多特征融合的肝细胞癌分化等级术前预测方法研究

A PREOPERATIVE PREDICTION METHOD FOR DIFFERENTIATION GRADES OF HEPATOCELLULAR CARCINOMA BASED ON MULTI-FEATURE FUSION

通过融合影像学特征和深度特征实现对肝细胞癌分化分级的无创术前预测。预测方法基于T2加权成像高通量提取影像学特征,使用SVM、随机森林、XGBoost和LightGBM等方法构造出影像学标签(Rad-score);利用EfficientNet-B7在增广图像数据上进行参数微调后提取了大量深度特征,并采用LightGBM构建深度标签(Deep-score);结合病人的临床特征利用回归模型构造诺模图进行可视化预测。实验结果证实模型具有较好的分类性能,最终分类模型AUC达到了0.828,校准曲线表现良好,可以为临床决策提供有价值的信息。

This paper realizes the non-invasive preoperative prediction of hepatocellular carcinoma differentiation level by fusing imaging features and depth features. The prediction method extracted the radiomics features based on T2 weighted imaging high-throughput. We used SVM, random forest, XGBoost and LightGBM to construct the Rad-score. By using EfficientNet-B7 to fine tune the parameters on the augmented image data, we extracted a large number of depth features. LightGBM was adopted to construct the Deep-score. Combining with the clinical characteristics of the patient, we used the regression model to build nomogram for visual prediction. The experimental results prove that the model has good prediction performance. The final model has an AUC of 0.828, and the calibration curve performs well, which can provide valuable information for clinical decision-making.