To Investigate the Prediction of Microvascular Invasion of Liver Cancer Based on MRI Texture and Image Fusion Model
ObjectieTo investigate the value of MRI texture analysis in preoperative prediction of microvascular invasion of liver cancer.Methods Retrospective analysis was performed on 102 patients with HCC confirmed by surgery and pathology in Affiliated Hospital of Chifeng University from January 2018 to June 2022.According to pathology,44 patients were classified into microvascular invasion MVI(+)group and 58 patients were classified into non-invasion MVI(-)group.Combined with MR Multi-sequence images,3D-Slicer was used to extract texture features of focus areas in MRI images.By statistical comparison and dimensionality reduction,the optimal MR Sequence was selected,and the prediction model was constructed based on the texture features and macro image features of the optimal sequence.Results One of the different clinical indicators between MVI(+)group and MVI(-)group was AFP,and there were 5 general image features.The mean mass size,unsmooth boundary,incomplete capsule,peritumoral enhancement,and low signal around hepatobiliary stage were closely associated with MVI invasion of HCC.After screening,the best sequence was portal phase image,and four best texture feature parameters related to MVI state were selected by LASSO logistic regression algorithm for dimensionality reduction,among which two were shape features and the other two were Baud signs.Texture features were combined with clinical data and general features to establish a fusion prediction model,which predicted that the AUC of MVI was 97％.Conclusion MRI texture analysis technique based on 3D-Slicer can accurately predict MVI before liver cancer.
NETL
NSTL
万方数据
维普
