Development and validation of a radiopathomics model for predicting liver metastases of colorectal cancer

To compare the ability of a model based on CT radiomics features, a model based on clinical data, and a fusion model based on a combination of both radiomics and clinical data to predict the risk of liver metastasis after surgery for colorectal cancer.

Two hundred and twelve patients with pathologically confirmed colorectal cancer were divided into a training set (n = 148) and a validation set (n = 64). Radiomics features from the most recent CT scans and clinical data obtained before surgery were extracted. Random forest models were trained to predict tumors with clinical data and evaluated using the area under the receiver-operating characteristic curve (AUC) and other metrics on the validation set.

Fourteen features were selected to establish the radiomics model, which yielded an AUC of 0.751 for the training set and an AUC of 0.714 for the test set. The fusion model, based on a combination of the radiomics signature and clinical data, showed good performance in both the training set (AUC 0.952) and the test set (AUC 0.761).

We have developed a fusion model that integrates radiomics features with clinical data. This fusion model could serve as a non-invasive, reliable, and accurate tool for the preoperative prediction of liver metastases after surgery for colorectal cancer.

Question Can a radiomics and clinical fusion model improve the prediction of liver metastases in colorectal cancer and help optimize clinical decision-making?

Findings The presented fusion model combining CT radiomics and clinical data showed superior accuracy in predicting colorectal cancer liver metastases compared to single models.

Clinical relevance Our study provides a non-invasive, relatively accurate method for predicting the risk of liver metastasis, improving personalized treatment decisions, and enhancing preoperative planning and prognosis management in colorectal cancer patients.