A CT-based radiomics nomogram for predicting histopathologic growth patterns of colorectal liver metastases

To develop a computed tomography (CT)-based radiomics nomogram for pre-treatment prediction of histopathologic growth patterns (HGPs) in colorectal liver metastases (CRLM) and to validate its accuracy and clinical value.

This retrospective study included a total of 197 CRLM from 92 patients. Lesions from CRLM were randomly divided into the training study (n = 137) and the validation study (n = 60) with the ratio of 3:1 for model construction and internal validation. The least absolute shrinkage and selection operator (LASSO) was used to screen features. Radiomics score (rad-score) was calculated to generate radiomics features. A predictive radiomics nomogram based on rad-score and clinical features was developed using random forest (RF). The performances of clinical model, radiomic model and radiomics nomogram were thoroughly evaluated by the DeLong test, decision curve analysis (DCA) and clinical impact curve (CIC) allowing for generation of an optimal predictive model.

The radiological nomogram model consists of three independent predictors, including rad-score, T-stage, and enhancement rim on PVP. Training and validation results demonstrated the high-performance level of the model of area under curve (AUC) of 0.86 and 0.84, respectively. The radiomic nomogram model can achieve better diagnostic performance than the clinical model, yielding greater net clinical benefit compared to the clinical model alone.

A CT-based radiomics nomogram can be used to predict HGPs in CRLM. Preoperative non-invasive identification of HGPs could further facilitate clinical treatment and provide personalized treatment plans for patients with liver metastases from colorectal cancer.