A PET-CT radiomics model for immunotherapy response and prognosis prediction in patients with metastatic colorectal cancer.

BACKGROUND

In recent years, radiomics, as a non-invasive method, has shown potential in predicting tumor response and prognosis by analyzing medical image data to extract high-dimensional features and reveal the heterogeneity of tumor microenvironment (TME).

OBJECTIVE

The aim of this study was to construct and validate a radiomic model based on PET/CT images for predicting immunotherapy response and prognosis in mCRC patients.

METHODS

This study included mCRC patients from multiple cohorts, including a training set (n=105), an internal validation set (n=60), a TME phenotype cohort (n=42), and an immunotherapy response cohort (n=99). High-dimensional radiomic features were extracted from PET/CT images using a deep neural network (DNN), and RNA-Seq was used to screen for features associated with TME phenotypes to construct a radiomic score (Rad-Score). At the same time, combined with immune scores (IHC staining results based on CD3 and CD8) and clinical features, a joint prediction model was developed to assess overall survival (OS) and progression-free survival (PFS). The predictive performance of the model was evaluated by area under receiver operating characteristic curve (AUC), calibration curve and decision curve analysis (DCA).

RESULTS

A radiomics signature to predict the TME phenotype was constructed in the training set and verified it in an internal validation set, with AUC of 0.855 and 0.844 respectively. In the TME phenotype external cohort, the radiomics signature can differentiate either immunopotentiation or immunosuppression tumor (AUC=0.814). In the immunotherapy response external cohort, the radiomics signature can predict response to immunotherapy (AUC=0.784). The combined nomograms can predict OS and PFS, with AUC of 0.860 and 0.875 respectively. The calibration curve and decision curve analysis (DCA) confirmed the predicting performance and clinical utility of the combined nomograms.

CONCLUSION

In this study, a radiomic model based on PET/CT images was successfully constructed, which can effectively predict immunotherapy response and prognosis of mCRC patients. The model combines radiomic features, immune scores and clinical features, showing high prediction accuracy and clinical application value. In the future, the reliability and generalization ability of this model need to be further verified in larger prospective studies to promote its application in clinical practice.