Identification and Prediction of Microvascular Invasion in Hepatocellular Carcinoma Based on Multiple Sequence Fusion Images

Hepatocellular carcinoma (HCC) is one of the most difficult diseases to be treated fundamentally in the medical field in today’s world, and the recurrence rate of HCC has been increasing year by year in recent years. The study on the identification and prediction of microvascular invasion phenomenon (MVI) in hepatocellular carcinoma is beneficial to the postoperative prevention of hepatocellular carcinoma patients, which can greatly improve the survival probability of patients after surgery. In this study, a new MVI prediction method (MSF-GWO-GPR) was proposed, which combines the multi-sequence image features of imaging histology, sifts out the unimportant features through the T-test method and the dimensionality reduction of Lasso regression, and establishes the multi-sequence Gaussian process regression (GPR) prediction model for data fusion, and combined with the Gray Wolf Optimization (GWO) algorithm to design a suitable optimization objective function to optimize the hyper-parameters in the GPR, in order to get better prediction results. Compared with the existing methods, the data from different sequences of MRI images are selected for fusion in this study, which makes the feature data richer and more complete, and the model applicability is better; moreover, by optimizing the hyperparameters in the model through the GWO, the model can be made more accurate, reduce the risk of misidentified prediction, and improve the accuracy rate. The experimental results show that MSF-GWO-GPR outperforms widely used machine learning methods in terms of accuracy, F1 score, precision, recall and AUC metrics. This may provide a new idea for the identification and prediction of MVI in hepatocellular carcinoma.