Habitat image based on enhanced MRI before transcatheter arterial chemoembolization of hepatocellular carcinoma clinical value of prognosis prediction

Objective

To explore the clinical value of predicting the 2-year progression-free survival time (PFS) in patients with hepatocellular carcinoma (HCC) before transcatheter arterial chemoembolization (TACE) based on enhanced MRI Habitat image model.

Methods

This study is a retrospective study that collected 114 HCC patients who received TACE treatment in the Interventional Clinic of Xinjiang Medical University from January 2022 to December 2023. The image was preprocessed (normalization, N4 calibration and image registration). The region of interest (ROI) of the registered image was sketched by ITK-SNAP software, and the characteristics of ROI were extracted with 19 characteristic parameters. Before the characteristics of the Habitat image, all pixels in the sketched ROI area should be analyzed by K-means clustering to get the best clustering number. Machine learn that obtained final characteristic parameter through a classifier algorithm to obtain a Habitat image model and an image omics model.

Results

Machine learning was carried out with classifiers SVM, MLP, KNN, LR and LightGBM, and the optimal classifier model parameters were selected to construct Habitat image model, imageology model and Habitat_Rad model. In the model, the diagnostic efficiency AUC value of classifier LightGBM was the highest, and the training cohort AUC value was 0.904, 0.847 and 0.925 respectively, and the validation set AUC value was 0.688 and 0.925 respectively. Habitat image segmentation is the best when K-means is 5, in which Habitat5 is usually located in the center of tumor, Habitat1 and Habitat4 are usually located in the periphery of tumor center, and Habitat2 and Habitat3 are usually located at the edge of tumor.

Conclusion

Comparison to existing prognostic models in HCC, Habitat_Rad is the best model to predict the survival rate of PFS after TACE, and it is considered that Habitat images can provide a new method to evaluate the spatial heterogeneity in HCC and become a new image biomarker in the future.