Prediction of post-hepatectomy liver failure risk in patients with colorectal cancer liver metastases utilizing gadoxetic acid-enhanced MRI radiomics with automatic segmentation and image registration.

Abstract

Objectives: To develop and validate a radiomics model based on gadoxetic acidenhanced MRI, which can better reflect the future liver remnant (FLR) and its function through automatic segmentation and image registration, thereby predicting the risk of post-hepatectomy liver failure (PHLF) in patients with colorectal cancer liver metastases (CRLM).

Methods: This retrospective study analyzed 134 CRLM patients who were divided into a training group (n = 93) and a validation group (n = 41). Model-_Whole was constructed using the entire liver on preoperative hepatobiliary phase (HBP) obtained by automatic segmentation excluding the region of liver metastases. Model-_FLR was constructed using the FLR on the HBP by registering postoperative CT and preoperative MR. Model-_Clinical was constructed using general characteristics, preoperative data, intraoperative data and imaging parameters by univariate and multivariate analyses. Model-_{Clinical-radiomics} was constructed using the optimal radiomics model and features from clinical model. The models were used to predict low-grade PHLF (without PHLF or PHLF Grade A) and high-grade PHLF (PHLF Grade B or C).

Results: There were three and five radiomics features used for constructing Model-_Whole and Model-_FLR. The number of resected lesions and international normalized ratio (INR) were used for constructing Model-_Clinical. The AUC of Model-_{Clinical-radiomics} was 0.912 (95%CI: 0.835-0.961) and 0.896 (95%CI: 0.760-0.969) in the training group and the validation group, higher than that of any other models.

Conclusion: Model-_{Clinical-radiomics}, which was based on Model-_FLR, had good efficacy in predicting PHLF.